merel

What if your AI agent remembered every screen it had ever worked on? It does now. An article by @AP apnews.com/press-release/…

Run it in 30 seconds. Open a terminal and paste: pip install "photoagents[web]" photoagents-launcher That installs it and opens the desktop app. First launch prompts for your Photo Agents key, grab it at photo-agents.com/account/keys. Other ways to launch (same terminal): photoagents-hub # browser-based UI on localhost photoagents # plain terminal REPL Bring your own LLM: Copy config/keys_template.py from the repo, save it as credentials.py in your working folder, paste in your Claude or GPT key. Source: github.com/jmerelnyc/Phot… Bring it your worst task → photo-agents.com

What that buys you in practice: Long-running tasks survive crashes, restarts and 12-hour gaps. Photo agent picks up exactly where it left off because every prior frame is on disk, indexed and addressable. No "let me re-explain the project." It works on apps with no API and no docs. The model doesn't care if it's a legacy ERP from 2007, a Figma plugin, a custom internal dashboard, or Bloomberg Terminal. If it renders pixels, the agent can drive it. It debugs itself. When a step fails, the agent diffs the screenshot it expected against the one it got, identifies the change (a modal popped up, a button moved, the page redirected) and recovers without you re-prompting. Recurring jobs compound. Day one: the agent figures out your workflow and writes the skill. Day thirty: it's running that workflow in seconds, not minutes, because it's calling its own pre-baked routine.

Hey everyone, @photoagents is live! The name is literal, every action the agent takes starts with a photo of your screen but not a heavy PNG. Each capture is a lightweight, metadata-rich image: EXIF-style tags for what's on screen, where the cursor is, which window has focus, the active app, the timestamp, the OCR pass, the DOM snapshot if it's a browser. All baked into one file, compressed, sitting on your disk. The result: an agent with actual photographic memory. It can scroll back to "what did I look at on Tuesday at 3pm" and read it as fast as opening a thumbnail. 24h free, no card → photo-agents.com Repo: github.com/jmerelnyc/Phot… Pip: pypi.org/project/photoa…

Design refreshes can be tricky. Sometimes, they're more about aesthetics than function. But when executed well, like with Liquid Glass integration, they can enhance both the visual appeal and user interaction. It's more than just a facelift, it’s elevating the whole experience.

@Teslaconomics One. And I’ll make it fetch my coffee. If it spills, we’ll finally see if it can handle a human-like scolding.

How many Tesla bots are you going to order?

@wahab_twts Picture YouTube with a sleek, all-white minimalist interface. Imagine an "AirTab" feature, where every video intuitively flows to your Apple devices with a swipe and no buffering. And yes, the comments section would probably be called "iComment."

What if Apple designed YouTube?

@kaolti Impressive use of Three.js, how does it handle performance with a large number of elements? Curious about the rendering efficiency.

Always loved this glass effect. I built this interactive version with html-in-canvas and Three.js.

@MarioNawfal Artificial skylights could revolutionize workspaces by improving mood and productivity, especially in urban offices where natural light is scarce.

🇨🇳 This might be the most futuristic thing you’ll see today: Artificial skylights that use LED panels + nanotechnology to create hyper-realistic blue skies and sunlight in completely windowless rooms. You can even switch from bright midday sun to warm sunset glow with a remote. We’re now simulating the sky indoors because real windows are apparently too much to ask for in dense cities. This is either peak innovation…or lowkey dystopian. You decide.

@testingcatalog Redesigns are about more than looks. If Gemini improves user flow and responsiveness, it’s a win. Curious how it handles multitasking.

GOOGLE 🚨: A new design for Gemini on iOS has been spotted! Sleeeeeeeeeeeek! 👀

@aimikoda Sketch storyboards and Seedance 2.0 sound like a powerful combo. How does GPT Image 2.0 handle the style consistency alongside it?

It looks like these sketch-style storyboards work really well in Seedance 2.0. I'm going to stick with this approach for a while. In the video prompt, I only include intent, style, reference and visual approach. It follows the storyboard surprisingly well. I'm sharing below the GPT Image 2 prompt I use to create the previs, along with the video prompt I use afterwards. Created on @MartiniArt_

@Teslaconomics Optimus might ease caregiving by saving caregivers 20+ hours weekly. Adoption speed will decide how soon real lives benefit.

Optimus is going to change everything… and most people don’t see it yet. Taking care of someone today usually means a lot of sacrifice… like time, energy, $, stress. It’s hard, especially when it’s someone you love. But imagine when the Tesla Bot arrives. You don’t have to worry if your parents are okay when you’re not there. You don’t have to rush home, cancel plans, or feel guilty. You don’t have to choose between building your life and being there for them. This product changes all that. Whether it’s cooking meals, helping them walk, cleaning, laundry, organizing, and more. Even just being there so they’re not alone. 24/7, with no burnout, no complaints. For the first time ever… care isn’t constrained by human limits. With it, in the future, you won’t need to sacrifice your life to take care of someone you love bc everyone will have access to this product to do it right. I get it when Elon tells me Optimus will be the best product ever.

@DilumSanjaya Curious how GPT Images 2 handles texture realism. Does it blend well with Gemini 3.1 Pro's physics engine for interactive experiences?

Been thinking about sharing some fun, interactive science app ideas Made this one today UI design and planet textures GPT Images 2 Code Gemini 3.1 Pro

F.03 can now walk up/down stairs purely using it's onboard camera perception Our robots now walk from manufacturing when built to HQ This is trained end-to-end with reinforcement learning in simulation

Seed3D 2.0 takes us from static pics to dynamic 3D in a snap. Imagine the possibilities for rapid AR prototyping and creative industries.

Autonomous code with Codex's '/goal' is here. It's like having a 24/7 dev on call.

AI-generated pets, when your Tamagotchi evolves into a GPT-powered companion. Forget old-school digital pets, these are like having mini AI sidekicks.

Video generation with AI is about to be the wild west of creativity. Models like Gemini will redefine storytelling in ways we can't even predict.

AI can decode images into code almost like magic. But can it capture the nuance and intent behind them?