On Saturday, crowds gathered in cities throughout the USA to protest President Donald Trump and his administration. Organizers of the No Kings rallies declare that greater than 7 million folks attended in all, throughout 2,700 cities within the Unites States and past. The gatherings offered a transparent image not solely of how widespread the resistance to the Trump administration has grow to be, but additionally the range of the coalition driving it. To not point out the indicators.
“In the present day, thousands and thousands of People stood collectively to reject authoritarianism and remind the world that our democracy belongs to the folks, to not one man’s ambition,” stated Ezra Levin and Leah Greenberg, cofounders of the nonprofit Indivisible—which took the lead on organizing the rallies—in a press release.
Forward of Saturday, Home speaker Mike Johnson described the deliberate gatherings because the “hate America rally” and warned that the crowds can be stuffed with “antifa varieties.” In actuality, the protests have been uniformly peaceable, with inflatable costumes showing to outnumber conservative bogeymen by a large margin. And Trump’s response to No Kings? An AI-generated video of himself, carrying a crown, piloting a fighter jet and dropping large quantities of excrement on protesting US residents under.
In the true world, the crowds walked their routes with out subject. Beneath are snapshots of No Kings from cities throughout the US, a have a look at a protest motion that’s more and more motivated and capable of mobilize.
The Counter Menace Unit™ (CTU) analysis group analyzes safety threats to assist organizations shield their methods. Primarily based on observations in July and August, CTU™ researchers recognized the next noteworthy points and adjustments within the world risk panorama:
Ransomware stays a unstable risk regardless of disruptions
Absent MFA permits exploitation of stolen credentials
Legacy vulnerabilities keep their worth
Ransomware stays a unstable risk regardless of disruptions
Regulation enforcement actions have made an affect on the ransomware ecosystem however haven’t diminished the variety of assaults.
Ransomware continues to pose a serious risk to organizations. Despite the fact that the variety of victims posted to leak websites has declined since reaching an all-time peak in March 2025, the figures in July and August remained increased than in the identical months of 2024. Regardless of the quantity of media consideration dedicated to high-profile ransomware and knowledge extortion assaults by Scattered Spider and ShinyHunters, the 2 most prolific schemes throughout July and August have been Qilin and Akira. Each schemes have been extremely energetic throughout 2025 (though the variety of victims posted every month is decrease than the month-to-month quantity from earlier prolific operations like LockBit). Even so, general, the excessive variety of ransomware assaults within the second and third quarters of 2025 have principally been extra evenly distributed throughout a number of teams in comparison with earlier years.
Regulation enforcement actions in opposition to main ransomware operators in 2024 and 2025 have precipitated fragmentation and volatility within the ransomware panorama. Sometimes, every legislation enforcement disruption conjures up a short lived spike in new group creations. Thirty-seven new schemes appeared within the first half of 2025. 4 others emerged in July, adopted by one other 4 in August. The sustained excessive quantity could also be linked to the common cadence of legislation enforcement exercise concentrating on LockBit all through 2024. A number of teams that had been thought of dormant returned to exercise throughout July and August 2025 too. In complete, 52 ransomware schemes have been energetic in August, a month-to-month quantity that has solely been surpassed 3 times within the earlier two years.
The brand new schemes are seemingly not fashioned by new cybercriminals. As an alternative, associates who labored with disrupted operations could be part of or launch a brand new scheme, both pooling efforts with different displaced associates or working independently. New schemes may additionally be rebrands of older operations. Associates could alternatively transfer to established operations corresponding to Akira that then improve their assault tempo because of having extra assets. These rebrands and circulation of associates throughout present or new teams could make kill chains laborious to determine and attribution tougher.
Despite the fact that these developments can improve the general issue of monitoring the ransomware ecosystem, they don’t significantly change the important thing defenses in opposition to most ransomware assaults: immediate patching, particularly of internet-facing gadgets; phishing-resistant multi-factor authentication (MFA); and complete monitoring of endpoints and networks. As well as, it’s turning into more and more necessary to observe cloud and hybrid environments for malicious exercise as risk actors pivot to the cloud.
What You Ought to Do Subsequent
Monitor authorities initiatives on advancing cloud safety.
Absent MFA permits exploitation of stolen credentials
Implementing MFA prevents risk actors from cashing in on stolen credentials.
CTU researchers have noticed a number of incidents the place cybercriminals or state-sponsored risk actors obtained preliminary entry to their sufferer’s atmosphere by abusing VPN credentials. For instance, the GOLD LEAPFROG risk group abused VPN credentials in an early 2025 assault that culminated within the deployment of SafePay ransomware.
Unauthorized entry of this nature permits risk actors to bypass conventional safety measures and achieve direct entry into inner methods, even when the equipment is totally patched in opposition to identified vulnerabilities. Different kinds of generally abused entry embody distant desktop logins or administrative accounts. In different phrases, strategies designed to guard approved entry for distant staff also can give attackers entry if the safety is just not sufficiently sturdy.
Menace actors usually buy the credentials on underground marketplaces. Infostealer malware steals credentials and different knowledge from methods that it infects. The stolen knowledge is then packaged into logs and bought on-line to different risk actors. Tens of millions of logs can be found on the market, and the quantity continues to rise sharply annually. Because of this, defending methods from infostealer infections kinds a key a part of defending in opposition to subsequent ransomware or knowledge extortion assaults.
Menace actors who acquire partial credentials may additionally attempt to brute-force entry on VPN accounts. In the event that they achieve gaining entry, it’s nearly all the time as a result of the VPN doesn’t require MFA to authenticate. MFA alone doesn’t stop all unauthorized entry, but it surely does cut back the specter of essentially the most opportunistic cybercrime. Implementing phishing-resistant MFA on all internet-facing providers and home equipment reduces threat ranges even additional. Such a MFA makes use of hardware-backed strategies to stop token theft.
What You Ought to Do Subsequent
Evaluate steering revealed by U.S. Cybersecurity and Infrastructure Safety Company (CISA) on implementing phishing-resistant MFA.
Legacy vulnerabilities keep their worth
Even when a vulnerability is years outdated, it’s not often too late to patch.
In August, the Federal Bureau of Investigation (FBI) warned that Russian state-sponsored risk actors linked to the Russian Federal Safety Service’s (FSB) Middle 16 have been conducting cyberespionage assaults in opposition to U.S. and different entities by actively concentrating on Cisco gadgets unpatched in opposition to a vulnerability from 2018. CTU researchers noticed related exercise by Russian state-sponsored risk actors in 2023.
The FBI was additionally one in every of a number of companies within the U.S. and past to concern a warning about Chinese language state-sponsored risk actors compromising networks worldwide for espionage functions. The a part of the doc that lined how the attackers gained preliminary entry states that “they’re having appreciable success exploiting publicly identified widespread vulnerabilities” slightly than beforehand unknown zero-day vulnerabilities. The doc lists the Cisco vulnerability from 2018, in addition to others from 2023 and 2024 that have an effect on edge gadgets.
Organizations could not patch promptly for a lot of causes. Budgetary restrictions and restricted personnel are simply two components that may affect a patching program. Employees could lack consciousness of the vulnerability or not understand that impacted tools is in use. Some patches could require extra analysis or have to be changed with workarounds to keep away from potential affect to different business-critical operations. In some instances, tools is so outdated that distributors now not publish safety updates. Nonetheless, unpatched edge gadgets put organizations in danger. Throughout incident response engagements that CTU researchers noticed in 2024, vulnerabilities in internet-facing gadgets have been the commonest preliminary entry vectors.
The danger posed by unpatched gadgets is just not going to say no. It’s already straightforward to make use of freely accessible scanning methods and publicly accessible exploit code to seek out and exploit susceptible methods, and it’s possible that AI may automate this additional. Immediate patching in keeping with enterprise threat calculations or substitute of end-of-life methods stay extra necessary than ever.
What You Ought to Do Subsequent
Monitor authorities and vendor advisories and different risk intelligence sources about risk actor conduct, and comply with patching recommendation as applicable in your atmosphere.
Conclusion
Regardless of adjustments in risk group composition and will increase in assault numbers, some features of the cyber risk stay the identical. Cybercriminals and state-sponsored risk actors proceed to reap the benefits of quick access to organizations’ environments. Luckily, the fundamentals of excellent cyber protection additionally stay fixed: immediate patching, phishing-resistant MFA, and complete monitoring and response.
We’ve all seen AI write essays, compose music, and even paint jaw-dropping portraits. But there may be one other frontier that’s far more thrilling – AI-generated movies. Think about stepping right into a film scene, sending an animated greeting, or witnessing a historic reenactment, all crafted by AI. Till now, most of us had been simply curious spectators, giving directions and hoping for the best output. However what if you happen to might transcend that and construct your personal video technology webapp?
That’s precisely what I did with Time Capsule. Right here is the way it works: you add a photograph, choose a time interval, choose a occupation, and similar to that, you might be transported into the previous with a customized picture and a brief video. Easy, proper? However the true magic occurred after I took this concept to the Knowledge Hack Summit, essentially the most futuristic AI convention in India.
We turned Time Capsule right into a GenAI playground sales space designed purely for enjoyable and engagement. It grew to become the favorite sales space not only for attendees, however for audio system and GenAI leaders too. Watching individuals’s faces gentle up as they noticed themselves as astronauts, kings, or Victorian-era students jogged my memory why constructing with AI is so thrilling.
So I believed, why not share this concept with the beautiful viewers of AnalyticsVidhya. Buckle up, as I take you behind the scenes of how Time Capsule went from an concept to an interactive video technology webapp.
The Idea of a Video Technology WebApp (With Time Capsule Instance)
At its core, a video technology webapp is any utility that takes consumer enter and transforms it into a brief, AI-created video. The enter might be a selfie, textual content, or a number of easy selections. The AI then turns them into shifting visuals that really feel distinctive and private.
Each video technology app works by means of three foremost blocks:
Enter: What the consumer gives – this might be a photograph, textual content, or picks.
Transformation: The AI interprets the enter and creates visuals.
Output: The ultimate outcome, delivered as a video (and typically a picture too).
The actual energy lies in personalization: Generic AI movies on-line might be enjoyable, however movies starring you immediately develop into extra partaking and memorable.
Ideas like Time Capsule thrive as a result of they don’t simply generate random clips, they generate your story, or on this case, your journey by means of time.
How Time Capsule Works
Right here is the brief and simple means during which Time Capsule, our video technology webapp, works.
Add a photograph of your self.
Choose ethnicity, time interval, occupation, and motion.
AI generates a customized portrait and brief video.
As soon as achieved, you obtain your personal time-travel expertise, whether or not as a Roman gladiator, a Renaissance artist, or perhaps a futuristic explorer.
Now that you simply’ve seen how the method works, it’s time to begin constructing your personal ‘Time Capsule’.
Applied sciences Utilized in TimeCapsule
Listed below are all of the applied sciences utilized in constructing our very personal video technology webapp – TimeCapsule.
Programming Language
Python: Core language for scripting the applying and integrating AI providers.
AI & Generative Fashions
OpenAI API: For enhancing prompts and producing text-based steerage for photographs and movies.
Google Gemini (genai): For picture evaluation (e.g., gender detection) and generative duties.
RunwayML: AI picture technology from prompts and reference photographs.
fal_client (FAL AI): Accessing Seeddance professional mannequin for video technology from a single picture and motion immediate.
Pc Imaginative and prescient
OpenCV (cv2): Capturing photographs from a webcam and processing video frames.
Pillow (PIL): Dealing with photographs, overlays, and including a emblem to movies.
NumPy: Array manipulation for photographs and frames throughout video processing.
E-mail Integration
Yagmail: Sending emails with attachments (generated picture and video).
Utilities & System
Requests: Downloading generated photographs and movies through HTTP requests.
uuid: Producing distinctive identifiers for information.
os: Listing creation, file administration, and atmosphere entry.
dotenv: Loading API keys and credentials from .env information.
datetime: Timestamping generated information.
base64: Encoding photographs for API uploads.
enum: Defining structured choices for ethnicity, time interval, occupation, and actions.
re: Sanitizing and cleansing textual content prompts for AI enter.
The best way to Make Your Personal Time Capsule
Now that you already know all the weather that make the Time Capsule attainable, right here is the precise, step-by-step blueprint to make your personal video-generation webapp.
1. Import All Libraries
You’ll first must import all crucial libraries for the challenge.
import cv2 import os import uuid import base64 import requests import yagmail import fal_client import numpy as np from PIL import Picture import google.generativeai as genai from enum import Enum from dotenv import load_dotenv from openai import OpenAI from datetime import datetime import time import re from runwayml import RunwayML # Load atmosphere variables load_dotenv()
2. Enter from Consumer
The method of the net app begins with the consumer importing a private photograph. This photograph types the muse of the AI-generated character. Customers then choose ethnicity, time interval, occupation, and motion, offering structured enter that guides the AI. This ensures the generated picture and video are customized, contextually correct, and visually partaking.
Seize Picture
The capture_image methodology makes use of OpenCV to take a photograph from the consumer’s digital camera. Customers can press SPACE to seize or ESC to cancel. It consists of fallbacks for instances when the digital camera GUI isn’t out there, routinely capturing a picture if wanted. Every photograph is saved with a novel filename to keep away from overwriting.
1. Initialize Digicam
Listed below are the steps to initialize the digital camera.
cap = cv2.VideoCapture(0)
Opens the default digital camera (gadget 0).
Checks if the digital camera is accessible; if not, prints an error and exits.
2. Begin Seize Loop
whereas True: ret, body = cap.learn()
Constantly reads frames from the digital camera.
ret is True if a body is efficiently captured.
The body incorporates the precise picture knowledge.
3. Show the Digicam Feed
Strive: cv2.imshow('Digicam - Press SPACE to seize, ESC to exit', body) key = cv2.waitKey(1) & 0xFF besides cv2.error as e: # If GUI show fails, use automated seize after delay print("GUI show not out there. Utilizing automated seize...") print("Capturing picture in 3 seconds...") time.sleep(3) key = 32 # Simulate SPACE key press
Reveals the reside digital camera feed in a window.
Waits for consumer enter: – SPACE (32) → Seize the picture. – ESC (27) → Cancel seize.
Fallback: If the GUI show fails (e.g., working in a headless atmosphere), the code waits 3 seconds and routinely captures the picture.
4. Save the Picture
unique_id = str(uuid.uuid4()) timestamp = datetime.now().strftime("%Ypercentmpercentd_percentHpercentMpercentS") filename = f"captured_{timestamp}_{unique_id}.jpg" filepath = os.path.be part of('captured_images', filename) # Save the picture cv2.imwrite(filepath, body)
Right here is the whole code to seize the picture, in a single piece:
def capture_image(self): """Seize picture utilizing OpenCV with fallback strategies""" print("Initializing digital camera...") cap = cv2.VideoCapture(0) if not cap.isOpened(): print("Error: Couldn't open digital camera") return None attempt: print("Digicam prepared! Press SPACE to seize picture, ESC to exit") whereas True: ret, body = cap.learn() if not ret: print("Error: Couldn't learn body") break # Attempt to show the body attempt: cv2.imshow('Digicam - Press SPACE to seize, ESC to exit', body) key = cv2.waitKey(1) & 0xFF besides cv2.error as e: # If GUI show fails, use automated seize after delay print("GUI show not out there. Utilizing automated seize...") print("Capturing picture in 3 seconds...") # import time time.sleep(3) key = 32 # Simulate SPACE key press if key == 32: # SPACE key # Generate UUID for distinctive filename unique_id = str(uuid.uuid4()) timestamp = datetime.now().strftime("%Ypercentmpercentd_percentHpercentMpercentS") filename = f"captured_{timestamp}_{unique_id}.jpg" filepath = os.path.be part of('captured_images', filename) # Save the picture cv2.imwrite(filepath, body) print(f"Picture captured and saved as: {filepath}") break elif key == 27: # ESC key print("Seize cancelled") filepath = None break besides Exception as e: print(f"Error throughout picture seize: {e}") # Fallback: seize with out GUI print("Making an attempt fallback seize...") attempt: ret, body = cap.learn() if ret: unique_id = str(uuid.uuid4()) timestamp = datetime.now().strftime("%Ypercentmpercentd_percentHpercentMpercentS") filename = f"captured_{timestamp}_{unique_id}.jpg" filepath = os.path.be part of('captured_images', filename) cv2.imwrite(filepath, body) print(f"Fallback picture captured and saved as: {filepath}") else: filepath = None besides Exception as e2: print(f"Fallback seize additionally failed: {e2}") filepath = None lastly: cap.launch() attempt: cv2.destroyAllWindows() besides: cross # Ignore if GUI cleanup fails return filepath
Output:
Select ethnicity, Time interval, Career, and Motion
The get_user_selections methodology permits customers to customise their character by selecting from the next choices: ethnicity, time interval, occupation, and motion. Choices are displayed with numbers, and the consumer inputs their selection. The picks are returned and used to create a customized picture and video.
Listed below are all of the choices out there to select from:
class EthnicityOptions(Enum): CAUCASIAN = "Caucasian" AFRICAN = "African" ASIAN = "Asian" HISPANIC = "Hispanic" MIDDLE_EASTERN = "Center Japanese" MIXED = "Blended Heritage" class TimePeriodOptions(Enum): Jurassic = "Jurassic Interval (200-145 million in the past)" ANCIENT = "Historical Occasions (Earlier than 500 AD)" MEDIEVAL = "Medieval (500-1500 AD)" RENAISSANCE = "Renaissance (1400-1600)" COLONIAL = "Colonial Period (1600-1800)" VICTORIAN = "Victorian Period (1800-1900)" EARLY_20TH = "Early twentieth Century (1900-1950)" MID_20TH = "Mid twentieth Century (1950-1990)" MODERN = "Fashionable Period (1990-Current)" FUTURISTIC = "Futuristic (Close to Future)" class ProfessionOptions(Enum): WARRIOR = "Warrior/Soldier" SCHOLAR = "Scholar/Trainer" MERCHANT = "Service provider/Dealer" ARTISAN = "Artisan/Craftsperson" FARMER = "Farmer/Agricultural Employee" HEALER = "Healer/Medical Skilled" ENTERTAINER = "Entertainer/Performer" NOBLE = "Noble/Aristocrat" EXPLORER = "Explorer/Adventurer" SPIRITUAL = "Religious Chief/Clergy" class ActionOptions(Enum): SELFIE = "Taking a selfie from digital camera view" DANCING = "Dancing to music" WORK_ACTION = "Performing work/skilled motion" WALKING = "Easy strolling" COMBAT = "Fight/preventing motion" CRAFTING = "Crafting/creating one thing" SPEAKING = "Talking/giving a speech" CELEBRATION = "Celebrating/cheering"
Right here is the code block to seize the choice:
def get_user_selections(self): """Get consumer picks for character customization""" print("n=== Character Customization ===") # Ethnicity choice print("nSelect Ethnicity:") for i, choice in enumerate(EthnicityOptions, 1): print(f"{i}. {choice.worth}") ethnicity_choice = int(enter("Enter selection (1-6): ")) - 1 ethnicity = checklist(EthnicityOptions)[ethnicity_choice] # Time Interval choice print("nSelect Time Interval:") for i, choice in enumerate(TimePeriodOptions, 1): print(f"{i}. {choice.worth}") period_choice = int(enter("Enter selection (1-9): ")) - 1 time_period = checklist(TimePeriodOptions)[period_choice] # Career choice print("nSelect Career:") for i, choice in enumerate(ProfessionOptions, 1): print(f"{i}. {choice.worth}") profession_choice = int(enter("Enter selection (1-10): ")) - 1 occupation = checklist(ProfessionOptions)[profession_choice] # Motion Choice print("n=== Video Motion Choice ===") for i, motion in enumerate(ActionOptions, 1): print(f"{i}. {motion.worth}") action_choice = int(enter("Choose motion (1-8): ")) - 1 action_choice = checklist(ActionOptions)[action_choice] return ethnicity, time_period, occupation,action_choice
Detect Gender from the Picture
The detect_gender_from_image perform makes use of Google Gemini 2.0 Flash to determine the gender from an uploaded picture. It handles errors gracefully, returning ‘particular person’ if detection fails. This helps personalize the generated video, making certain the mannequin precisely represents the consumer and avoids producing a male picture for a feminine or vice versa.
def detect_gender_from_image(self, image_path): """Detect gender from captured picture utilizing Google Gemini 2.0 Flash""" attempt: print("Analyzing picture to detect gender...") # Add picture to Gemini uploaded_file = genai.upload_file(image_path) # Anticipate the file to be processed # import time whereas uploaded_file.state.title == "PROCESSING": print("Processing picture...") time.sleep(2) uploaded_file = genai.get_file(uploaded_file.title) if uploaded_file.state.title == "FAILED": print("Didn't course of picture") return 'particular person' # Generate response response = self.gemini_model.generate_content([ uploaded_file, "Look at this image and determine if the person appears to be male or female. Respond with only one word: 'male' or 'female'." ]) # Clear up the uploaded file genai.delete_file(uploaded_file.title) gender = response.textual content.strip().decrease() if gender in ['male', 'female']: return gender else: return 'particular person' # fallback besides Exception as e: print(f"Error detecting gender with Gemini: {e}") return 'particular person' # fallback
3. Generate an Picture from the Inputs
Now that now we have the enter from the consumer for all of the parameters, we will proceed to creating an AI picture utilizing the identical. Listed below are the steps for that:
Generate a Immediate for Picture Technology
After accumulating the consumer’s picks, we use the enhance_image_prompt_with_openai perform to create an in depth and interesting immediate for the picture technology mannequin. It transforms the essential inputs like gender, ethnicity, occupation, time interval, and motion right into a inventive, skilled, and age-appropriate immediate, making certain the generated photographs are correct, visually interesting, and customized.
For this, we’re utilizing the “gpt-4.1-mini” mannequin with a temperature of 0.5 to introduce some randomness and creativity. If the OpenAI service encounters an error, the perform falls again to a easy default immediate, holding the video technology course of clean and uninterrupted.
def enhance_image_prompt_with_openai(self, ethnicity, time_period, occupation, gender,motion): """Use OpenAI to reinforce the picture immediate based mostly on consumer picks""" base_prompt = f""" Create a easy, clear immediate for AI picture technology: - Gender: {gender} - Ethnicity: {ethnicity.worth} - Career: {occupation.worth} - Time interval: {time_period.worth} - Performing Motion: {motion.worth} - Present applicable clothes and setting - Make the background a bit distinctive within the immediate - Preserve it applicable for all ages - Most 30 phrases """ attempt: response = self.openai_client.chat.completions.create( mannequin="gpt-4.1-mini", messages=[{"role": "user", "content": base_prompt}], max_tokens=80, temperature=0.5 ) enhanced_prompt = response.selections[0].message.content material.strip() return enhanced_prompt besides Exception as e: print(f"Error with OpenAI: {e}") # Fallback immediate return f"{gender} {ethnicity.worth} {occupation.worth} from {time_period.worth} performing {motion.worth}, skilled portrait"
After producing a immediate, we have to clear and sanitize it for API compatibility. Right here is the perform for sanitizing the immediate.
def sanitize_prompt(self, immediate): """Sanitize and restrict immediate for API compatibility""" # Take away problematic characters and restrict size import re # Take away further whitespace and newlines immediate = re.sub(r's+', ' ', immediate.strip()) # Take away particular characters which may trigger points immediate = re.sub(r'[^ws,.-]', '', immediate) # Restrict to 100 phrases most phrases = immediate.cut up() if len(phrases) > 100: immediate=" ".be part of(phrases[100]) # Guarantee it is not empty if not immediate: immediate = "Skilled portrait {photograph}" return immediate
Generate Knowledge URI for the Picture
The image_to_data_uri perform converts a picture right into a Knowledge URI, permitting it to be despatched instantly in API requests or embedded in HTML. It encodes the file as Base64, detects its sort (JPEG, PNG, or GIF), and creates a compact string format for seamless integration.
def image_to_data_uri(self, filepath): """Convert picture file to knowledge URI for API""" with open(filepath, "rb") as image_file: encoded_string = base64.b64encode(image_file.learn()).decode('utf-8') mime_type = "picture/jpeg" if filepath.decrease().endswith(".png"): mime_type = "picture/png" elif filepath.decrease().endswith(".gif"): mime_type = "picture/gif" return f"knowledge:{mime_type};base64,{encoded_string}"
Generate Picture utilizing RunwayML
As soon as now we have generated the Immediate and Knowledge URI of the Picture. Now its time for the AI to do its magic. We are going to use runwayML to generate Picture. You need to use completely different picture technology device out there out there.
The perform generate_image_with_runway is answerable for producing a picture utilizing RunwayML.
Import and initialize RunwayML
from runwayml import RunwayML runway_client = RunwayML()
It hundreds the RunwayML library and creates a consumer to work together with the API.
Put together the immediate
print(f"Utilizing immediate: {immediate}") print(f"Immediate size: {len(immediate)} characters") # Sanitize immediate yet another time immediate = self.sanitize_prompt(immediate)
The immediate supplied by the consumer is printed and cleaned (sanitized) to make sure it doesn’t break the mannequin.
Convert reference picture
data_uri = self.image_to_data_uri(image_path)
The enter picture is transformed right into a Knowledge URI (a Base64 string) so it may be handed to RunwayML as a reference.
Generate picture with RunwayML
process = runway_client.text_to_image.create( mannequin="gen4_image", prompt_text=immediate, ratio='1360:768', reference_images=[{ "uri": data_uri}] ).wait_for_task_output()
It sends the sanitized immediate + reference picture to RunwayML’s gen4_image mannequin to generate a brand new picture.
As soon as RunwayML returns a URL, it downloads the picture. A novel filename (based mostly on time and UUID) is created, and the picture is saved within the intermediate_images folder.
Error dealing with & fallback
If one thing goes flawed with the principle immediate, the perform retries with an easier immediate (simply ethnicity + occupation).
If even that fails, it returns None.
Right here the entire Code Block for Picture Technology:
Now that now we have generated a picture based mostly precisely on the consumer enter, right here is the method to transform this picture to a video.
Generate a Immediate for Video Technology
The enhance_video_prompt_with_openai perform turns consumer selections into protected, inventive video prompts utilizing GPT-4.1-mini. It adapts tone based mostly on occupation i.e., critical for warriors, gentle and humorous for others, whereas holding content material family-friendly.
To take care of consistency, it additionally ensures the character’s face stays the identical throughout the video. Together with consumer picks, the picture technology immediate is handed too, so the video has full context of the character and background. If OpenAI fails, a fallback immediate retains issues working easily.
def enhance_video_prompt_with_openai(self, motion, image_prompt, ethnicity, time_period, occupation, gender): """Enhanced video immediate technology - simplified and protected""" if occupation == "WARRIOR": video_prompt_base = f""" Context from picture immediate : {image_prompt} Get Context from picture immediate and generate an in depth and protected video immediate: - Character: A {gender} {ethnicity.worth} {occupation.worth} - Motion: {motion.worth} - Time Interval: {time_period.worth} - Focus fully on the motion. - Preserve the language easy and applicable. - Scene needs to be life like. - Keep away from controversial subjects or violence. - Video needs to be applicable for all ages. """ else: video_prompt_base = f""" Context from picture immediate : {image_prompt} Get Context from picture immediate and generate a easy, protected and humorous video immediate: - Character: A {gender} {ethnicity.worth} {occupation.worth} - Motion: {motion.worth} - Time Interval: {time_period.worth} - Focus fully on the motion. - Preserve the language easy and applicable. - Make it little bit humorous. - Scene needs to be life like and humorous. - Keep away from controversial subjects or violence. - Video needs to be applicable for all ages """ attempt: response = self.openai_client.chat.completions.create( mannequin="gpt-4.1-mini", messages=[{"role": "user", "content": video_prompt_base}], max_tokens=60, temperature=0.5 ) enhanced_video_prompt = response.selections[0].message.content material.strip() enhanced_video_prompt = enhanced_video_prompt+" Preserve face of the @particular person constant in entire video." return enhanced_video_prompt besides Exception as e: print(f"Error enhancing video immediate: {e}") # Fallback immediate - quite simple return f"{gender} {ethnicity.worth} {occupation.worth} from {time_period.worth} performing {motion.worth}, skilled video"
Generate Video with Seedance V1 Professional
For video technology, we’re utilizing the Seedance V1 Professional mannequin. To entry this mannequin, we’re utilizing fal.ai. Fal AI gives entry to Seedance Professional at a less expensive value. I’ve examined many video technology fashions like Veo3, Kling AI, and Hailuo. I discover Seedance finest for this objective because it has a lot better face consistency and is less expensive. The one downside is that it doesn’t present audio/music within the video.
def generate_video_with_fal(self, image_path, video_prompt): """Generate video utilizing fal_client API with error dealing with""" attempt: print("Producing video with fal_client...") print(f"Utilizing video immediate: {video_prompt}") # Add the generated picture to fal_client print("Importing picture to fal_client...") image_url = fal_client.upload_file(image_path) print(f"Picture uploaded efficiently: {image_url}") # Name the mannequin with the uploaded picture URL print("Beginning video technology...") outcome = fal_client.subscribe( "fal-ai/bytedance/seedance/v1/professional/image-to-video", arguments={ "immediate": video_prompt, "image_url": image_url, "decision": "720p", "length": 10 }, with_logs=True, on_queue_update=self.on_queue_update, ) print("Video technology accomplished!") print("Outcome:", outcome) # Extract video URL from outcome if outcome and 'video' in outcome and 'url' in outcome['video']: video_url = outcome['video']['url'] print(f"Video URL: {video_url}") # Obtain the video print("Downloading generated video...") response = requests.get(video_url) if response.status_code == 200: # Generate distinctive filename for video unique_id = str(uuid.uuid4()) timestamp = datetime.now().strftime("%Ypercentmpercentd_percentHpercentMpercentS") video_filename = f"generated_video_{timestamp}_{unique_id}.mp4" video_filepath = os.path.be part of('final_videos', video_filename) # Save the video with open(video_filepath, "wb") as file: file.write(response.content material) print(f"Video generated efficiently: {video_filepath}") return video_filepath else: print(f"Didn't obtain video. Standing code: {response.status_code}") return None else: print("No video URL present in outcome") print("Full outcome construction:", outcome) return None besides Exception as e: print(f"Error producing video with fal_client: {e}") if "delicate" in str(e).decrease(): print("Content material flagged as delicate. Attempting with an easier immediate...") # Strive with very fundamental immediate basic_prompt = "particular person shifting" attempt: image_url = fal_client.upload_file(image_path) outcome = fal_client.subscribe( "fal-ai/bytedance/seedance/v1/professional/image-to-video", arguments={ "immediate": basic_prompt, "image_url": image_url, "decision": "720p", "length": 10 }, with_logs=True, on_queue_update=self.on_queue_update, ) if outcome and 'video' in outcome and 'url' in outcome['video']: video_url = outcome['video']['url'] response = requests.get(video_url) if response.status_code == 200: unique_id = str(uuid.uuid4()) timestamp = datetime.now().strftime("%Ypercentmpercentd_percentHpercentMpercentS") video_filename = f"generated_video_{timestamp}_{unique_id}.mp4" video_filepath = os.path.be part of('final_videos', video_filename) with open(video_filepath, "wb") as file: file.write(response.content material) print(f"Video generated with fundamental immediate: {video_filepath}") return video_filepath besides Exception as e2: print(f"Even fundamental immediate failed: {e2}") return None return None
Output:
Add your emblem within the video (optionally available):
In case you’re making a video to your group, you may simply add a watermark to it. This helps defend your content material by stopping others from utilizing the video for industrial functions.
The add_logo_to_video perform provides a emblem watermark to a video. It checks if the brand exists, resizes it, and locations it within the bottom-right nook of each body. The processed frames are saved as a brand new video with a novel title. If one thing goes flawed, it skips the overlay and retains the unique video.
def add_logo_to_video(self, video_path, logo_width=200): """Add emblem overlay to video earlier than emailing""" attempt: print("Including emblem overlay to video...") # Verify if emblem file exists if not os.path.exists(self.logo_path): print(f"Emblem file not discovered at {self.logo_path}. Skipping emblem overlay.") return video_path # Load emblem with transparency utilizing Pillow emblem = Picture.open(self.logo_path).convert("RGBA") # Open the enter video cap = cv2.VideoCapture(video_path) # Get video properties width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) peak = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) fps = cap.get(cv2.CAP_PROP_FPS) fourcc = cv2.VideoWriter_fourcc(*'mp4v') # Create output filename unique_id = str(uuid.uuid4()) timestamp = datetime.now().strftime("%Ypercentmpercentd_percentHpercentMpercentS") output_filename = f"video_with_logo_{timestamp}_{unique_id}.mp4" output_path = os.path.be part of('final_videos', output_filename) out = cv2.VideoWriter(output_path, fourcc, fps, (width, peak)) # Resize emblem to specified width logo_ratio = logo_width / emblem.width logo_height = int(emblem.peak * logo_ratio) emblem = emblem.resize((logo_width, logo_height)) frame_count = 0 total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) whereas True: ret, body = cap.learn() if not ret: break # Present progress frame_count += 1 if frame_count % 10 == 0: progress = (frame_count / total_frames) * 100 print(f"Processing body {frame_count}/{total_frames} ({progress:.1f}%)") # Convert body to PIL Picture frame_pil = Picture.fromarray(cv2.cvtColor(body, cv2.COLOR_BGR2RGB)).convert("RGBA") # Calculate place (backside proper nook with padding) pos = (frame_pil.width - emblem.width - 10, frame_pil.peak - emblem.peak - 10) # Paste the brand onto the body frame_pil.alpha_composite(emblem, dest=pos) # Convert again to OpenCV BGR format frame_bgr = cv2.cvtColor(np.array(frame_pil.convert("RGB")), cv2.COLOR_RGB2BGR) out.write(frame_bgr) cap.launch() out.launch() print(f"Emblem overlay accomplished: {output_path}") return output_path besides Exception as e: print(f"Error including emblem to video: {e}") print("Persevering with with unique video...") return video_path
Ship this Video through E-mail
As soon as the video is generated, customers will wish to view and obtain it. This perform makes it attainable by sending the video on to the e-mail deal with they supplied.
def send_email_with_attachments(self, recipient_email, image_path, video_path): """Ship e-mail with generated content material utilizing yagmail""" attempt: # Get e-mail credentials from atmosphere variables sender_email = os.getenv('SENDER_EMAIL') sender_password = os.getenv('SENDER_PASSWORD') if not sender_email or not sender_password: print("E-mail credentials not present in atmosphere variables") return False yag = yagmail.SMTP(sender_email, sender_password) topic = "Your AI Generated Character Picture and Video" physique = f""" Whats up! Your AI-generated character content material is prepared! Connected you will discover: - Your generated character picture - Your generated character video (with emblem overlay) Thanks for utilizing our AI Picture-to-Video Generator! Greatest regards, AI Generator Crew """ attachments = [] if image_path: attachments.append(image_path) if video_path: attachments.append(video_path) yag.ship( to=recipient_email, topic=topic, contents=physique, attachments=attachments ) print(f"E-mail despatched efficiently to {recipient_email}") return True besides Exception as e: print(f"Error sending e-mail: {e}") return False
At this stage, you’ve constructed the core engine of your internet app, producing a picture, making a video, including a emblem, and sending it to the consumer through e-mail. The subsequent step is to attach all of it collectively by growing the frontend and backend for the net app.
Challenges
Constructing a customized video technology internet app comes with a number of technical and operational challenges:
1. Dealing with AI Failures and API Errors
AI fashions for picture and video technology can fail unexpectedly.
APIs could return errors or produce undesired outputs.
Fallback methods had been important to make sure clean operation, reminiscent of utilizing simplified prompts or different technology strategies.
2. Managing Delicate Content material
AI-generated content material can inadvertently produce inappropriate outcomes.
Implementing checks and protected prompts ensured that each one outputs remained family-friendly.
3. Consumer Expectations for Personalization
Customers count on extremely correct and customized outcomes.
Making certain gender, ethnicity, occupation, and different particulars had been appropriately mirrored required cautious immediate design and validation.
4. Finalizing the Video Technology Mannequin
Discovering a mannequin that maintained face consistency at an inexpensive price was difficult.
After testing a number of choices, Seedance V1 Professional through Fal.ai supplied the very best stability of high quality, consistency, and value.
5. Faux or Unreliable Aggregators
Selecting a dependable mannequin supplier was tough. Fal.ai labored nicely, however earlier I experimented with others:
Replicate.com: Quick however restricted customization choices, and later confronted fee points.
Pollo.ai: Good interface, however their API service was unreliable; it generated no movies.
The important thing takeaway: keep away from pretend or unreliable suppliers; all the time take a look at completely earlier than committing.
6. Time Administration and Efficiency
Video technology is time-consuming, particularly for real-time demos.
Optimizations like LRU caching and a number of API situations helped cut back latency and enhance efficiency throughout occasions.
Past the Time Capsule: What Else Might You Construct?
The Time Capsule is only one instance of a customized video technology app. The core engine might be tailored to create quite a lot of modern purposes:
Customized Greetings: Generate birthday or pageant movies that includes family and friends in historic or fantasy settings.
Advertising & Branding: Produce promotional movies for companies, including logos and customised characters to showcase services or products.
Academic Content material: Carry historic figures, scientific ideas, or literature scenes to life in a visually partaking means.
Interactive Storytelling: Permit customers to create mini-movies the place characters evolve based mostly on consumer enter, selections, or actions.
Gaming Avatars & Animations: Generate customized in-game characters, motion sequences, or brief cutscenes for sport storytelling.
The probabilities are countless, any situation the place you need customized, visible, and interactive content material, this engine may also help carry concepts to life.
Conclusion
The Time Capsule internet app reveals simply how far AI has come, from producing textual content and pictures to creating customized movies that really really feel like your personal. You begin with a easy photograph, choose a time interval, occupation, and motion, and in moments, the AI brings your historic or fantasy self to life. Alongside the best way, we deal with challenges like AI errors, delicate content material, and time-consuming video technology with good fallbacks and optimizations. What makes this thrilling isn’t simply the know-how, it’s the countless potentialities.
From enjoyable customized greetings to instructional storytelling, advertising movies, or interactive mini-movies, this engine might be tailored to carry numerous inventive concepts to life. With slightly creativeness, your Time Capsule might be the beginning of one thing actually magical.
Technical content material strategist and communicator with a decade of expertise in content material creation and distribution throughout nationwide media, Authorities of India, and personal platforms
Login to proceed studying and luxuriate in expert-curated content material.
Cleanlab is data-model and data-framework agnostic, a robust facet of its design. It doesn’t matter in the event you’re operating PyTorch, OpenAI, scikit-learn, or Tensorflow; Cleanlab can work with any classifier. It does, nonetheless, have particular workflows for frequent duties like token classification, multi-labeling, regression, picture segmentation and object detection, outlier detection, and so forth. It’s value perusing the instance set to see for your self how the method works and what outcomes you may count on.
Snakemake
Information science workflows are exhausting to arrange, and that’s even tougher to do in a constant, predictable means. Snakemake was created to automate the method, organising information evaluation workflows in ways in which guarantee everybody will get the identical outcomes. Many present information science initiatives depend on Snakemake. The extra transferring elements you’ve got in your information science workflow, the extra possible you’ll profit from automating that workflow with Snakemake.
Snakemake workflows resemble GNU Make workflows—you outline the steps of the workflow with guidelines, which specify what they absorb, what they put out, and what instructions to execute to perform that. Workflow guidelines may be multithreaded (assuming that provides them any profit), and configuration information may be piped in from JSON or YAML recordsdata. You too can outline features in your workflows to rework information utilized in guidelines, and write the actions taken at every step to logs.
Computational neuroscientist and singer/songwriter Kimaya (Kimy) Lecamwasam, who additionally performs electrical bass and guitar, says music has been a core a part of her life for so long as she will be able to keep in mind. She grew up in a musical household and performed in bands all by means of highschool.
“For many of my life, writing and taking part in music was the clearest approach I needed to categorical myself,” says Lecamwasam. “I used to be a extremely shy and anxious child, and I struggled with talking up for myself. Over time, composing and performing music turned central to each how I communicated and to how I managed my very own psychological well being.”
Together with equipping her with invaluable expertise and experiences, she credit her ardour for music because the catalyst for her curiosity in neuroscience.
“I acquired to see firsthand not solely the ways in which audiences reacted to music, but additionally how a lot worth music had for musicians,” she says. “That shut connection between making music and feeling properly is what first pushed me to ask why music has such a strong maintain on us, and finally led me to check the science behind it.”
Lecamwasam earned a bachelor’s diploma in 2021 from Wellesley Faculty, the place she studied neuroscience — particularly within the Methods and Computational Neuroscience monitor — and likewise music. Throughout her first semester, she took a category in songwriting that she says made her extra conscious of the connections between music and feelings. Whereas learning at Wellesley, she participated within the MIT Undergraduate Analysis Alternatives Program for 3 years. Working within the Division of Mind and Cognitive Sciences lab of Emery Brown, the Edward Hood Taplin Professor of Medical Engineering and Computational Neuroscience, she targeted totally on classifying consciousness in anesthetized sufferers and coaching brain-computer interface-enabled prosthetics utilizing reinforcement studying.
“I nonetheless had a extremely deep love for music, which I used to be pursuing in parallel to all of my neuroscience work, however I actually wished to attempt to discover a technique to mix each of these issues in grad college,” says Lecamwasam. Brown really useful that she look into the graduate applications on the MIT Media Lab inside the Program in Media Arts and Sciences (MAS), which turned out to be a perfect match.
“One factor I actually love about the place I’m is that I get to be each an artist and a scientist,” says Lecamwasam. “That was one thing that was essential to me once I was selecting a graduate program. I wished to make it possible for I used to be going to have the ability to do work that was actually rigorous, validated, and essential, but additionally get to do cool, artistic explorations and truly put the analysis that I used to be doing into apply in numerous methods.”
Exploring the bodily, psychological, and emotional impacts of music
Knowledgeable by her years of neuroscience analysis as an undergraduate and her ardour for music, Lecamwasam targeted her graduate analysis on harnessing the emotional efficiency of music into scalable, non-pharmacological psychological well being instruments. Her grasp’s thesis targeted on “pharmamusicology,” taking a look at how music may positively have an effect on the physiology and psychology of these with anxiousness.
The overarching theme of Lecamwasam’s analysis is exploring the assorted impacts of music and affective computing — bodily, mentally, and emotionally. Now within the third yr of her doctoral program within the Opera of the Future group, she is at the moment investigating the affect of large-scale reside music and live performance experiences on the psychological well being and well-being of each viewers members and performers. She can also be working to clinically validate music listening, composition, and efficiency as well being interventions, together with psychotherapy and pharmaceutical interventions.
Her current work, in collaboration with Professor Anna Huang’s Human-AI Resonance Lab, assesses the emotional resonance of AI-generated music in comparison with human-composed music; the goal is to determine extra moral functions of emotion-sensitive music era and suggestion that protect human creativity and company, and may also be used as well being interventions. She has co-led a wellness and music workshop on the Wellbeing Summit in Bilbao, Spain, and has introduced her work on the 2023 CHI convention on Human Elements in Computing Methods in Hamburg, Germany and the 2024 Audio Largely convention in Milan, Italy.
Lecamwasam has collaborated with organizations close to and much to implement real-world functions of her analysis. She labored with Carnegie Corridor’s Weill Music Institute on its Nicely-Being Concert events and is at the moment partnering on a examine assessing the affect of lullaby writing on perinatal well being with the North Shore Lullaby Undertaking in Massachusetts, an offshoot of Carnegie Corridor’s Lullaby Undertaking. Her predominant worldwide collaboration is with an organization known as Myndstream, engaged on initiatives evaluating the emotional resonance of AI-generated music to human-composed music and pondering of scientific and real-world functions. She can also be engaged on a undertaking with the businesses PixMob and Empatica (an MIT Media Lab spinoff), centered on assessing the affect of interactive lighting and large-scale reside music experiences on emotional resonance in stadium and enviornment settings.
Constructing group
“Kimy combines a deep love for — and complex data of — music with scientific curiosity and rigor in ways in which characterize the Media Lab/MAS spirit at its greatest,” says Professor Tod Machover, Lecamwasam’s analysis advisor, Media Lab college director, and director of the Opera of the Future group. “She has lengthy believed that music is likely one of the strongest and efficient methods to create customized interventions to assist stabilize emotional misery and promote empathy and connection. It’s this identical need to determine sane, secure, and sustaining environments for work and play that has led Kimy to turn out to be one of the vital efficient and devoted community-builders on the lab.”
Lecamwasam has participated within the SOS (College students Providing Assist) program in MAS for a couple of years, which assists college students from a wide range of life experiences and backgrounds in the course of the technique of making use of to the Program in Media Arts and Sciences. She is going to quickly be the primary MAS peer mentor as a part of a brand new initiative by means of which she’s going to set up and coordinate applications together with a “buddy system,” pairing incoming grasp’s college students with PhD college students as a approach to assist them transition into graduate pupil life at MIT. She can also be a part of the Media Lab’s Studcom, a student-run group that promotes, facilitates, and creates experiences meant to deliver the group collectively.
“I feel every thing that I’ve gotten to do has been so supported by the chums I’ve made in my lab and division, in addition to throughout departments,” says Lecamwasam. “I feel everyone seems to be simply actually excited concerning the work that they do and so supportive of each other. It makes it in order that even when issues are difficult or troublesome, I’m motivated to do that work and be part of this group.”
Information group warns Half 108 might impinge on free speech
By DRONELIFE Options Editor Jim Magill
A coalition of virtually two dozen main information media organizations has raised issues that the proposed FAA Half 108 rule might hamper journalists’ efforts to report the information utilizing drones.
In its remark to the FAA’s proposed rule, which might set up a standardized framework for drone operators to conduct common BVLOS flights, the Information Media Coalition (NMC), mentioned the rule as proposed might as an alternative “impede the vast majority of information organizations from partaking in these operations, that are a mechanism that permits free speech.”
The NMC’s statements have been among the many greater than 3,000 feedback the FAA acquired in response to its discover of proposed rulemaking for the Half 108 rule. The remark interval closed on October 6.
In its principal objection to the proposed rule, the Coalition mentioned that as presently written the rule would prohibit newsgathering BVLOS flights to areas with inhabitants densities of Class 3 – areas of average inhabitants, resembling developments and single-family houses — or decrease. This is able to preclude the information media from conducting BVLOS drone flights in city areas and on the websites of many main disasters involving giant teams of individuals, resembling hurricanes or floods.
Such restrictions would “exclude important geographic areas from BVLOS newsgathering, with out regard for whether or not such regulation is important to guard public security,” the Coalition’s remark states.
The Coalition additionally objected to a bit of the proposed rule that might restrict a UAS operator to flying not more than 24 “energetic” BVLOS-enabled drones. Such a restriction “might impede giant media firms with quite a few associates and freelance journalists, with none indication that such a strict restrict will improve security.”
As well as, the Coalition known as into query the necessity for newsgathering organizations to file detailed flight plans with the FAA previous to conducting BVLOS operations.
“Whereas the Coalition acknowledges that getting superior approval may be possible for some BVLOS operations, requiring journalists to submit detailed, superior flight plans would considerably undermine the business’s capacity to collect information,” the group’s remark mentioned. “To state the apparent, the information doesn’t all the time behave in a predictable format or remoted geographic grid, and journalists have to be permitted to react and modify in actual time.”
The FAA is below a good timeline for promulgating a ultimate BVLOS rule. In June, President Trump issued an Government Order requiring that the FAA finalize the Half 108 rule by early 2026. By legislation the FAA should assessment each remark to the proposed rule earlier than issuing a ultimate rule. The assessment course of is more likely to be additional difficult by the federal government shutdown, which started on October 1 and which reveals no indicators of reaching a conclusion any time quickly.
Inhabitants-density restrictions current a thorny downside for media
In an interview, Charles Tobin, a Washington D.C.-based lawyer with the agency of Ballard Spahr who ready the Coalition’s remark, mentioned he’s particularly involved in regards to the low-population density restrictions that the Half 108 order would impose on newsgathering.
“Journalists cowl information the place it occurs; in crowded areas, in rural areas and in areas in between,” he mentioned. “And so, to restrict us basically to rural areas is to not pay sufficient consideration to an vital end-user just like the information media, who function within the public curiosity.”
Tobin, who has represented the NMC on drone-related points since 2015, mentioned the information media has had a stellar security file for safely flying drones in pursuit of the information for the previous decade.
“We want the FAA to take one other take a look at the place the information media may have the ability to function and never restrict us to such a low-density space,” he mentioned. Tobin added that the media’s requirement for working BVLOS flights are comparatively minor in contrast with some industries that use drones extra extensively of their operations.
“We simply must get round that tree or that pole or that impediment the place we don’t have a visible line of sight,” he mentioned. “We’re not speaking about typically flying miles away from the operator. We’re speaking about additional dozens of yards at most in your typical news-gathering circumstance.”
Tobin additionally objected to the proposed restriction on the variety of drones that an organization and its associates might function at one time, saying it doesn’t make sense for giant broadcasting networks which may have giant numbers of affiliate stations and contractors who’re licensed to fly drones, who function not directly on behalf of the dad or mum firm.
“Are you able to think about telling an organization that has 100 tv stations, that just one quarter of your stations, even when they’re in very completely different elements of the nation, can fly?” he requested. “It serves no public security function, and it inhibits us from telling the information in our communities in the way in which that we do.”
He additionally mentioned getting FAA’s approval of pre-developed flight plans may make sense for different drone-using industries, however not for the information enterprise.
“There are particular information tales that do lend themselves to extra of a planning course of. However for essentially the most half, journalism is the protection of stories that breaks and when it occurs, it’s just about unattainable to have the ability to put collectively the kind of detailed flight planning that the FAA is considering,” he mentioned.
He added that the requirement that information operations file flight plans with the FAA might put a crimp in using drones in investigative reporting. “Lots of journalism — investigative journalism specifically — solely works if the topic of the journalism doesn’t know upfront that you simply’re going to be working, that you simply’re going to be protecting them, that you simply’re going to be watching them,” he mentioned.
Tobin mentioned the NMC has labored throughout the FAA’s public remark course of to attempt to short-circuit the implementation of a BVLOS rule that would impinge on press freedom.
“We’re a First Modification-protected business that often resists authorities management however we’ve got leaned into the method with the FAA as a result of we acknowledge that we should be a protected operator and a superb citizen,” he mentioned. “And we hope the FAA takes that under consideration in revising the principles to provide us a way more First Modification-friendly regime than is being proposed.”
Learn extra:
Jim Magill is a Houston-based author with nearly a quarter-century of expertise protecting technical and financial developments within the oil and gasoline business. After retiring in December 2019 as a senior editor with S&P International Platts, Jim started writing about rising applied sciences, resembling synthetic intelligence, robots and drones, and the methods during which they’re contributing to our society. Along with DroneLife, Jim is a contributor to Forbes.com and his work has appeared within the Houston Chronicle, U.S. Information & World Report, and Unmanned Programs, a publication of the Affiliation for Unmanned Automobile Programs Worldwide.
That’s how Meta’s Chief AI Scientist Yann LeCun described the blowback after OpenAI researchers did a victory lap over GPT-5’s supposed math breakthroughs.
Google DeepMind CEO Demis Hassabis added, “that is embarrassing.”
The Decoder experiences that in a since-deleted tweet, OpenAI VP Kevin Weil declared that “GPT-5 discovered options to 10 (!) beforehand unsolved Erdős issues and made progress on 11 others.” (“Erdős issues” are well-known conjectures posed by mathematician Paul Erdős.)
Nevertheless, mathematician Thomas Bloom, who maintains the Erdos Issues web site, mentioned Weil’s publish was “a dramatic misrepresentation” — whereas these issues had been certainly listed as “open” on Bloom’s web site, he mentioned that solely means, “I personally am unaware of a paper which solves it.”
In different phrases, it’s not correct to say GPT-5 was capable of resolve beforehand unsolved issues. As an alternative, Bloom wrote, “GPT-5 discovered references, which solved these issues, that I personally was unaware of.”
Sebastien Bubeck, an OpenAI researcher who’d additionally been touting GPT-5’s accomplishments, then acknowledged that “solely options within the literature had been discovered,” however he recommended this stays an actual accomplishment: “I understand how laborious it’s to go looking the literature.”
My Home windows 11 laptop computer is usually secure, nevertheless it sometimes encounters issues once I set up new {hardware} or apps with compatibility points. It could additionally turn into gradual when operating too many apps concurrently, and like all software program, it generally experiences unexplained random glitches. None of those are vital points, and I can often repair them myself in just some minutes utilizing the troubleshooting instruments constructed into Home windows 11.
Troubleshooting instruments, like System Supervisor and Occasion Viewer, are easy, and permit anybody to resolve widespread {hardware} and software program points while not having to take their laptop computer to a restore store. Even the extra superior troubleshooting instruments, just like the Command Immediate, are straightforward to study with a easy search and might resolve some Home windows 11 issues in minutes. I take advantage of these Home windows troubleshooting instruments to forestall giving strangers entry to my private data and keep away from paying for pointless repairs.
Activity Supervisor
Establish the useful resource hog
The Activity Supervisor is the primary place I look if my laptop computer is operating slowly or if I encounter an unresponsive app. This instrument is accessible utilizing CTRL+Alt+Del, and it shows an inventory of all of the apps presently operating on my laptop computer, together with the quantity of reminiscence, CPU, and disk area every one makes use of. It additionally gives a complete of all of the used options on the prime, in order that I can know the standing of my system.
The Activity Supervisor is the primary place I look if my laptop computer is operating slowly or if I encounter an unresponsive app.
If my laptop computer is sluggish, it is often as a result of background apps are losing my reminiscence, and I can manually shut them within the Activity Supervisor to unencumber sources. The Activity Supervisor may also point out if an app is unresponsive and let me shut it with out restarting my laptop computer. I discover the Startups app utility throughout the Activity Supervisor helpful for disabling packages that run mechanically and decelerate my laptop computer. I additionally use the Efficiency tab for monitoring system units in real-time to determine bottlenecks and different points which may be slowing my system down.
Occasion Viewer
Discover system errors
Occasion Viewer is my go-to instrument for locating why sure apps could also be freezing or working incorrectly. This app comprises logs of all my laptop computer’s purposes, system, and safety occasions, permitting me to determine potential {hardware} and software program conflicts on my laptop computer. It shows the supply of the issue, together with an outline, the time, and the severity degree, to simply troubleshoot the problems.
I sometimes create customized views to show the related data with out being overwhelmed by irrelevant particulars when utilizing the Occasion Viewer. I favor viewing logs from the final 24 hours and have set the app to determine errors, warnings, and demanding system logs. As soon as I decide which apps are producing the errors, I can try to restore or reinstall them utilizing the opposite instruments listed right here.
System Restore
Return your laptop computer to a earlier state
Most of us have skilled points after putting in an replace or app that wreaks havoc with our machines. System Restore is among the finest instruments for repairing these sorts of issues as a result of it could possibly return my laptop computer to an earlier state earlier than I put in the troublesome software program. This instrument is not enabled by default in Home windows 10 and 11, and I wanted to set it up manually and select which drives it protects.
I can create new restore factors if I am attempting new software program or adjusting my laptop computer’s settings.
As soon as enabled, I can create new restore factors if I am attempting new software program or adjusting my laptop computer’s settings. The app may also mechanically create restore factors if I make software program modifications. I can then use it to return my laptop computer to a working situation by selecting a restore level from the checklist. You’ll be able to’t interrupt the method (which often takes some time), so I let it run once I’m not busy and need not work on my laptop computer. It is nonetheless price utilizing as a result of it could possibly mechanically determine and resolve points that will take hours to seek out and restore manually.
System Supervisor
Discover faults and replace drivers
The System Supervisor is the perfect place to start out if my ports, sound, or different {hardware} aren’t functioning correctly. This easy troubleshooting instrument shows an inventory of all of the {hardware} on my laptop computer, together with show adapters, cameras, and different units. It’ll often show a yellow exclamation mark if it detects a {hardware} drawback or an unknown system, permitting me to determine the fault at a look with out losing time. I can double-click on every system to examine its standing and confirm that it’s working correctly earlier than trying to reinstall it or replace the drivers.
Units usually fail as a consequence of software program points, and the System Supervisor options built-in instruments to scan for {hardware} modifications, examine driver particulars, or replace drivers. The Replace Driver characteristic is useful as a result of it mechanically searches for the perfect driver and installs it. If it would not work, I can at all times use the Roll Again Driver operate to undo the modifications and restore the older model earlier than the replace. I will at all times use the Home windows System Supervisor to determine and rectify {hardware} points earlier than contemplating taking it to a restore middle that may do the identical factor and cost me for it.
Command Immediate
Repair startup points
The Command Immediate is a text-based interface embedded in most Home windows variations, permitting customers to entry information and run troubleshooting instruments. This instrument permits me to bypass the usual Home windows graphical consumer interface and run instructions to execute scripts or navigate to particular directories. It is precious for repairing corruptions and different errors that might forestall Home windows from booting appropriately. I additionally use it for copying knowledge off corrupt drives earlier than reinstalling the working system.
Most Home windows laptops can boot instantly into the Command Immediate, and it is price attempting these straightforward fixes in case your OS is experiencing startup points.
This troubleshooting instrument is not as user-friendly because the others listed right here and requires data of particular text-based instructions to operate appropriately. These instructions will be simply discovered on-line and used to scan and restore corrupt information (sfc /scannow), examine the drive for errors (chkdsk), and carry out different duties. Most Home windows laptops can boot instantly into the Command Immediate, and it is price attempting these straightforward fixes in case your OS is experiencing startup points.
In his Energy On e-newsletter at the moment, Bloomberg‘s Mark Gurman stated a few of Apple’s software program engineers have “issues” in regards to the overhauled Siri’s efficiency. Nonetheless, he didn’t present any particular particulars in regards to the shortcomings.
iOS 26.4 will probably be launched in March or April, so Apple nonetheless has round 5 to 6 months to enhance any lingering points with the voice assistant.
When it first previewed the brand new Siri at WWDC 2024, Apple stated it could have three most important capabilities: on-screen consciousness, private context, and in-app controls. For instance, Apple confirmed an iPhone person asking Siri about their mom’s flight and lunch reservation plans, based mostly on data from the Mail and Messages apps.
The brand new Siri was initially anticipated to debut with iOS 18.4 final 12 months, however Apple delayed it. In July, Apple CEO Tim Prepare dinner stated his firm was making “good progress” on the extra customized Siri, and he stated the options had been on monitor to launch subsequent 12 months. Nonetheless, if points persist, then even an iOS 26.4 launch won’t be assured.
Apple plans to chop manufacturing of the iPhone Air amid underwhelming gross sales efficiency, Japan’s Mizuho Securities believes (through The Elec).
The Japanese funding banking and securities agency claims that the iPhone 17 Professional and iPhone 17 Professional Max are seeing increased gross sales than their predecessors throughout the identical interval final 12 months, whereas the usual iPhone 17 is a significant success, performing…
iOS 26 was launched final month, however the software program prepare by no means stops, and iOS 26.1 beta testing is already underway. To this point, iOS 26.1 makes each Apple Intelligence and Dwell Translation on suitable AirPods out there in extra languages, and it consists of another minor modifications throughout the Apple Music, Calendar, Photographs, Clock, and Safari apps.
Extra options and modifications will observe in future …
Apple’s software program engineers proceed to internally take a look at iOS 26.0.2, in response to MacRumors logs, which have been a dependable indicator of upcoming iOS variations.
iOS 26.0.2 might be a minor replace that addresses bugs and/or safety vulnerabilities, however we have no idea any particular particulars but.
The replace will probably be launched by the tip of subsequent week.
Final month, Apple launched iOS 26.0.1,…
Apple on Wednesday up to date the 14-inch MacBook Professional, iPad Professional, and Imaginative and prescient Professional with its next-generation M5 chip, however earlier rumors have indicated that the corporate nonetheless plans to announce a minimum of a couple of extra merchandise earlier than the tip of the 12 months.
The next Apple merchandise have at one level been rumored to be up to date in 2025, though it’s unclear if the timeframe for any of them has…
Apple on Wednesday up to date the 14-inch MacBook Professional base mannequin with an M5 chip, and there are two key storage-related upgrades past that chip bump.
First, Apple says the brand new 14-inch MacBook Professional gives as much as 2× sooner SSD efficiency than the equal previous-generation mannequin, so learn and write speeds ought to get a major increase. Apple says it’s utilizing “the newest storage know-how,” …
Apple plans to launch MacBook Air fashions geared up with the brand new M5 chip in spring 2026, in response to Bloomberg’s Mark Gurman. Apple can also be engaged on M5 Professional and M5 Max MacBook Professional fashions that can come early within the 12 months.
Neither the MacBook Professional fashions nor the MacBook Air fashions are anticipated to get design modifications, with Apple specializing in easy chip upgrades. Within the case of the MacBook Professional, a m…
Whereas the brand new iPad Professional’s headline characteristic is the M5 chip, the system has another modifications, together with N1 and C1X chips, sooner storage speeds, and extra.
With the M5 chip, the brand new iPad Professional has as much as a 20% sooner CPU and as much as a 40% sooner GPU in comparison with the earlier mannequin with the M4 chip, in response to Geekbench 6 outcomes. Take into account that 256GB and 512GB configurations have a 9-core CPU,…
The primary alleged benchmark end result for the M5 chip within the new 14-inch MacBook Professional has surfaced, permitting for some efficiency comparisons.
Primarily based on a single unconfirmed end result uploaded to the Geekbench 6 database at the moment, the M5 chip has pulled off a formidable feat. Particularly, the chip achieved a rating of 4,263 for single-core CPU efficiency, which is the best single-core rating that…
iOS 26.4 is anticipated to introduce a revamped model of Siri powered by Apple Intelligence, however not everyone seems to be happy with how effectively it really works.
In his Energy On e-newsletter at the moment, Bloomberg’s Mark Gurman stated a few of Apple’s software program engineers have “issues” in regards to the overhauled Siri’s efficiency. Nonetheless, he didn’t present any particular particulars in regards to the shortcomings.
iOS 26.4 will…
Testing out the COROS APEX 4 for the previous couple of weeks has given me flashbacks to sporting the 2022 Garmin Forerunner 955, my favourite operating watch earlier than Garmin and its rivals started shifting away from MIP shows. Loads of health watch followers detest this development, like my 955-wearing brother-in-law. Whereas I will admit I choose AMOLED shows, the APEX 4 is a nice look ahead to the holdouts.
COROS’s final watch, the NOMAD, challenges the Garmin Intuition 3 in a slender class: reasonably priced, rugged mountaineering watches with weeks of battery life. An MIP show felt like the correct match.
The APEX 4 targets path runners and “mountain athletes” who additionally care about battery life and out of doors visibility. However its $429/479 worth offers it stiffer competitors from the Suunto Race 2, Garmin Forerunner 570, and Polar Vantage M3 — all AMOLED watches, and the primary two with 2,000 nits for glorious readability in direct daylight.
Preliminary ideas
(Picture credit score: Michael Hicks / Android Central)
I reviewed the COROS APEX 2 and APEX 2 Professional in 2022 once I was nonetheless comparatively new to wearables. The {hardware} impressed me, particularly the epic battery life, however I might inform that the GPS and HR accuracy could possibly be higher, the well being monitoring and software program had loads of gaps, and the MIP shows have been painfully dim.
Unsurprisingly, the APEX 4 is extra reliably correct (as you may see under), the battery life nonetheless blows me away, and the MIP show has significantly better distinction, making it serviceable — although for no matter cause, the NOMAD’s show is extra readable indoors. I believe the APEX 4’s sapphire layer catches extra obscuring mild.
Most significantly, the APEX 4’s new Ambiq Apollo 510 processor powers the speediest, smoothest maps I’ve seen outdoors of Apple and Put on OS watches, the place most health watch rivals like Garmin are laggy. You now get avenue and path names, and the crown makes it seamless to zoom in to a complicated path flip or out to a wider view in a few seconds.
Picture 1 of 2
(Picture credit score: Michael Hicks / Android Central)
(Picture credit score: Michael Hicks / Android Central)
My solely complaints are with the turn-by-turn navigation for downloaded programs. It largely works with out situation, however I observed twice that once I got here to a loop, the arrow recommended I’m going in the wrong way that I selected once I made the course.
Get the most recent information from Android Central, your trusted companion on this planet of Android
After I deliberately went astray, the APEX 4 caught it and warned me instantly. However the turn-by-turn instructions do not replace primarily based on which route you are going through, so once I missed my proper flip and circled, it nonetheless advised me to show proper. I hope COROS can handle each of my map nitpicks sooner or later.
With its new processor, the APEX 4 is primed for years of upgrades. Over three years, the APEX 2 added HRV and stress information, sleep high quality, and operating kind evaluation, Strava Stay Segments, and security alerts, amongst dozens of latest options. Whereas Garmin watch updates are likely to dry up after 1–2 years, APEX 4 updates ought to final for much longer.
I particularly anticipate its built-in depth gauge sensor to allow scuba diving in some unspecified time in the future, and COROS advised Chase the Summit that it might use the mic & speaker to allow offline voice instructions sooner or later.
COROS APEX 4 GPS accuracy
All of the close by satellite tv for pc alerts (Picture credit score: Michael Hicks / Android Central)
Not all dual-band GPS watches are created equal. Having examined almost each smartwatch model’s GPS accuracy, I’ve discovered that COROS’s most up-to-date watches come the closest to Garmin for dependability, with the APEX 4 following that sample. Prepare for some boringly straight GPS traces, with solely a few glitches to talk of.
I took three dual-frequency GPS watches — the Garmin Forerunner 970, Google Pixel Watch 4, and APEX 4 — out for a 19-mile run; whereas Google relies upon primarily on L1 and L5 GPS frequencies, Garmin and COROS additionally pull in different GNSS alerts as wanted.
Picture 1 of 7
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
The COROS APEX 4 (orange) and Garmin Forerunner 970 (blue)(Picture credit score: Android Central)
By the top, Garmin measured 0.01 miles extra and Google 0.01 miles much less; you possibly can learn my Pixel Watch 4 health check if you wish to see its outcomes, however I’ve targeted on the 2 predominant rivals right here. By and enormous, all nineteen miles confirmed the 2 alerts both overlapping or in shut parallel with one another. They tended to ding concurrently or inside seconds at every mile marker.
I solely have one criticism: The APEX 4’s remaining map confirmed me warping from my present spot to a earlier one, then again, as seen above. It did not have an effect on my distance outcomes, nevertheless it did add about 50m to my remaining altitude totals. I observed an identical glitch mountaineering with the NOMAD in August, including 170m of additional ascent however no additional distance.
Picture 1 of 5
(Picture credit score: Android Central)
(Picture credit score: Android Central)
(Picture credit score: Android Central)
(Picture credit score: Android Central)
(Picture credit score: Android Central)
For a second 8-mile run, fortunately, we’re proper again to fantastically, dully correct GPS information from each the APEX 4 and Forerunner 970, their GPS traces overlapping or paralleling each other for nearly all the run. I’ve additionally observed that the APEX 4 tends to do barely higher with tunnels and underpasses, staying straight whereas Garmin’s line waves barely.
Picture 1 of 7
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
The COROS APEX 4 (blue) and Garmin Forerunner 970 (orange)(Picture credit score: Android Central)
Throughout a 5-mile hike, neither watch was good, every having moments the place they strayed barely from the satellite tv for pc path path. Each ended with the identical 5.44-mile distance, whereas my Suunto Race 2 measured 5.43 miles.
What’s odd is that COROS measured about 100 ft of additional elevation acquire (1,762 ft) in comparison with Garmin (1,660 ft) and Suunto (1,644 ft). I am undecided what to make of two actions with extra ascent totals, so I will hold testing it in my full APEX 4 overview.
COROS APEX 4 HR accuracy
(Picture credit score: Michael Hicks / Android Central)
I’ve typically discovered COROS watches’ HR accuracy to be dependable, however not good, particularly for the very best anaerobic information. I do not thoughts as a result of the COROS HRM is reasonably priced and cozy to enhance your information, and most COROS watches are low-cost, however you’d need the pricier APEX 4 to face by itself deserves.
(Picture credit score: Android Central)
This graph exhibits how the COROS APEX 4 and Garmin Forerunner 970 optical HR sensors evaluate. COROS and Garmin’s HR averages ended 1 bpm aside, however you possibly can see how the 2 watches react to my adjustments in effort equally quick and usually mirror each other; the one distinction is that COROS’s outcomes have been barely larger.
(Picture credit score: Android Central)
A extra dependable accuracy check got here throughout my 19-mile run, throughout which I wore a Garmin HRM 200 chest strap as a management group. The COROS APEX 4 ended with the identical 161 bpm common and 178 bpm most price.
You’ll be able to definitely see factors within the graph the place the 2 diverge as I climbed into larger coronary heart charges. The APEX 4 trails behind the chest strap to hit larger HRs, then stays excessive whereas the chest strap measures my effort degree lowering.
(Picture credit score: Michael Hicks / Android Central)
The picture above, with the Forerunner synced to my chest strap, exhibits the way you won’t instantly see how arduous you are pushing with the APEX 4 at any given second. So should you used the APEX 4 for an anaerobic monitor exercise, I might anticipate the HR common would fall a few bpm brief. That is frequent for wrist-based optical sensors, however hold that in thoughts.
(Picture credit score: Android Central)
Wrist-based optical sensors also can battle with sustained, decrease HR ranges, however the COROS APEX 4 matched the Garmin HRM 200 chest strap for HR common and most throughout my two-hour hike. It is intently aligned 90% of the time, with the APEX 4 doing very effectively at catching when my HR rose on robust hills. However there have been a number of moments the place my effort dipped however the APEX 4 stayed 5–10 bpm above my HR.
Is the COROS APEX 4 the correct look ahead to you?
(Picture credit score: Michael Hicks / Android Central)
COROS reps mentioned throughout our APEX 4 briefing that they “will nonetheless have a break up of AMOLED and MIP watches sooner or later,” however that they “nonetheless discover MIP to be an exceptional answer.” The one present COROS AMOLED is the wonderful COROS PACE Professional, however I do not assume there can be an APEX 4 Professional with AMOLED anytime quickly.
With out wading into the MIP vs. AMOLED debate, the APEX 4 is the correct match for individuals who prioritize out of doors visibility and longer battery life. I might argue the most recent 2,000-nit Garmin watches are completely seen outdoor, however their battery life does endure for the additional brightness, whereas the APEX 4 has that basic longevity, to the purpose that I nonetheless have not charged it since I unboxed it.
(Picture credit score: Michael Hicks / Android Central)
The APEX 4 jogs my memory of premium health watches from three years in the past, when there was much less business stress so as to add “smarts” such as you’d discover on an Apple or Galaxy Watch. The UI revolves solely round health: What’s subsequent in your marathon coaching plan, how a lot coaching load, have you ever constructed up, and the way does your HRV, stress, and sleep play into your restoration?
However I am much less tolerant of sure issues in 2025 than I used to be in 2022, like how the essential COROS UI crams dozens of instruments into an annoying, rotating Toolbox, or the overly-large “COROS” label subsequent to the show that detracts from its premium look.
Selecting or skipping the APEX 4 comes right down to your priorities as an athlete. Switching to the Forerunner 570, for instance, would internet you a flashlight, every day run ideas, higher music apps, contactless funds, and (arguably) a greater show, however would lose you sapphire glass, a extra fashionable metal bezel, offline maps, and 5–11 days of battery life.
The APEX 4 is what grumbling Garmin customers say that they need: A return to the MIP days, the place they targeted on health necessities and battery life over costly bells and whistles, with an affordable worth and no irritating feature-locking of one of the best options like maps.
I will have a full overview coming quickly with extra function testing, nevertheless it’s undoubtedly one of many greatest operating watches of 2025…should you do not choose AMOLED.
The COROS APEX 4 gives a 1.3-inch AMOLED contact show with sapphire glass, 41 hours of dual-band GPS with a 24-day battery life general, 32GB of storage for music, maps, and routes, and a mic & speaker for Bluetooth calls and Voice Pin diary entries throughout your hikes, that are then saved to your actions to share with mates. It is constructed for severe out of doors athletes.
