I had 14 tabs open just to keep up with AI. arXiv, Papers With Code, every leaderboard, HuggingFace, half a dozen RL-env hubs... So I built one screen for all of it. The Bloomberg terminal for AI research. It's called Sophon 🧵

it's time to change the scope of my X, as for last 4 months I've been working not only on Gen AI but mostly on building agentic systems that help people to run Meta ads. Meanwhile check out the product that I've been working on for last month: anrie.ai/

👀

Generating awesome toys of myself with a single photo using new image-to-3D model by @MeshyAI 1. Generate toy image with your photo using nano banana (prompt on image) 2. Get the image and generate a 3D model .threejs using Meshy 6 on @fal

We won $10k GPU with this project on a hackathon by @bfl_ml and @aiengine_hack last week, here is our workflow: 1. You upload a floor plan -> generating top-down and side views of furnished and decorated flat w/ nano banana. 2. Based on this 3D model is generated (Trellis on @fal) 3. Using top down and side view we generate 360 equirectangular images using nano banana -> embed them using panellum.js 4. Room editing with a prompt: FLUX-2 Pro or Nano banana. That's only a part of a stack though, lot's of tricky heuristics hiding underneath!

Luma Ray 3.14 just dropped by @LumaLabsAI and I compared it with Ray 3, released earlier this month - the quality jump is INSANE! Workflow tip: edit videos straight with a prompt. Check examples prompts:

Visited @veedstudio office today — met @sab8a and Aleksei, was super excited to meet the team behind VEED Fabric! Talked about the future of agentic video editing, avatar model improvements, and ways to level up the product even more. Looking forward to more collabs! 🚀

got sucked into the Clawdbot rabbit hole too, lol running it on my old MacBook Pro in terminal, connected to Telegram, caffeinate to keep it awake added custom ElevenLabs Scribe API for better voice transcription — now I voice-note tasks and it manages my Obsidian notes 🙈

We just won the @bfl_ml hack 🚀 Walked away with an $11k GPU — what’s the first thing you’d do with it? Surreal to meet @andi_blatt, whose FLUX got me into GenAI in the first place. Huge shoutout to @aiengine_hack and @zqwq333!

Building today at @bfl_ml BFL FLUX 2 and @nvidia hackathon in London @aiengine_hack

playing around with remotion skills for claude code. Looks really cool, but requires lots of prompting and style guiding, so I'm not finished here! Current prompt history in gist: gist.github.com/hellesgrind/…

Just discovered insane Claude Code hack - you can explicitly ask to launch any of X sub-agents for your task. Claude Code tends to save on token usage so it only launches 2-3 agents. This way you can dramatically enhance your coding workflow.

Back in days when we were starting the Unreal Labs - our first idea was to create automated AI tiktok channels, created a bunch but left them abandoned - we've found a pmf with AI ads. 1.5 years passed and it turns out our accounts became quite popular: > couple of M views total!

Some friday evening AI slop, have a great weekend!

working on podcast generator, and what you're building today

testing out this awesome claude interview approach shared by @trq212 add this to your claude md, describe your task to claude and ask to "interview me". It will follow up with LOTS of questions and then will write a plan that you can copy to new session and one-shot a new feature! Prompt for claude md: interview me in detail using the AskUserQuestionTool about literally anything: technical implementation, UI & UX, concerns, tradeoffs, etc. but make sure the questions are not obvious be very in-depth and continue interviewing me continually until it's complete, then write the spec to the file

Anthropic just dropped Claude Cowork - now you don't need Claude Code to work with your files. Currently in preview mode but I have early access. used it for writing recap of my Obsidian Notes and it's awesome!

Just made myself dance in the a-ha "Take on Me" style using Mixed Media by @higgsfield This is so damn cool

Checked out the new Mixed Media feature by @higgsfield—adding effects to videos is sooo easy now. Workflow is simple: 1. Go to @higgsfield_ai Open Mixed Media 2. Drop your video 3. Pick FPS style Results look pretty wild! Will post some more examples soon🫣

Sharing workflow for these miniature figure images 1. Go to Google AI Studio -> select nano banana pro 2. Upload your selfie prompt For prompt ask ChatGPT to write different prompts using this template: Make a miniature, full-body, isometric, realistic figurine of this person, wearing ABC, doing XYZ, on a white background, minimal, 4K resolution