Robin Rombach

🔥 I am so damn excited to announce the launch of Black Forest Labs. We set ourselves on a mission to advance state-of-the-art, high-quality generative deep learning models for images and video, and make them available to the broadest audience possible. Today, we release FLUX.1

@giffmana Macron tribute panel

haha love the GTC panel "black sunglass + black dress" trend, you can see who likes having fun:

The @bfl_ml team released Klein KV and showed how KV-caching can incorporated in a flow pipeline 🤯 The idea is simple and elegant. In the first denoising step, reference image tokens are included in the full DiT forward pass. Their per-layer KVs are computed and cached. In the subsequent steps, KVs for only noisy latents are computed while the cached reference KVs are injected during computing attention. As a result, it delivers upto 2.5x speedups for multi-reference editing tasks over Klein. I basically learned about it from this PR: github.com/huggingface/di… The PR is a poetry in motion and is from the BFL team itself! Kudos to them for always being the best when it comes to designing codebases for flow and diffusion models. The best! Check out the model here: huggingface.co/black-forest-l…

My favorite editing model, FLUX.2 [klein] 9B, just got 2x faster: Meet FLUX.2 [klein] 9B-KV 😍💨 > Using KV-Cache Optimization to reduce computation & speed up inference by up to 2.5 times for multi-reference editing love how well it edits "around" the bullets

@seungwookh This is so fucking cool 🥹

Can language models learn useful priors without ever seeing language? We pre-pre-train transformers on neural cellular automata — fully synthetic, zero language. This improves language modeling by up to 6%, speeds up convergence by 40%, and strengthens downstream reasoning. Surprisingly, it even beats pre-pre-training on natural text! Blog: hanseungwook.github.io/blog/nca-pre-p… (1/n)

🤖🤖🤖🌲🌲🌲

@CSProfKGD @ELLISforEurope Awesome, see you there!

Just confirmed that I will be speaking at the @ELLISforEurope Winter School on Foundation Models (FoMo) in Amsterdam later this month 🤗

@natanielruizg Very cool!

Excited to show some surprising inventions on generative multiplayer games we made at Google with Stanford. We call the work MultiGen. I've always been inspired by early studios like id Software with Doom or Blizzard with Warcraft bringing networked video games to the next level. We are at the point in history where we can make strides like them, but for generative games. It's a strange feeling to be in the age of generative video games while still discovering how exactly to train the models and design the tools that make them useful. All of the tools that have been invented for classic game engines need to be redesigned for generative games. For example level and world design is not entirely possible with existing technology. We introduce editable memory to diffusion game engines that allow for design of new levels via a minimap. But we can easily imagine how this can be expanded with different creation tools. The end goal of this research direction is to allow game designers to be able to guide the generation process of their world, at the granularity that they prefer. Editable memory also allows us to add multiplayer to Generative Doom. We were amazed when we saw GameNGen some years ago, and now you can play it live with friends in real-time, on your couch or even online. Shared representations like our editable memory seem like the future for this type of experience. Models are, in some cases, expensive and approximate encoders but great interpolators and extrapolators. Leveraging their strengths lets you have completely new experiences that can be realized now and not in the distant future. This work was started at my previous team and continued in collaboration with Stanford. Congratulations to all for the discoveries.

@RuiqiGao 🌲🌲🌲

🚀🚀🚀

New paper out! We present a training method for multimodal generative models, called Self-Flow, which combines classic flow matching and representation learning. Why? Unlike most representation alignment methods, our new approach does not require external, pretrained models and thus scales gracefully to joint multimodal training on images, videos and audio. How? It combines per-timestep flow matching with dual-timestep representation learning, improving the models' internal representations. This approach outperforms prior methods and shows promising scaling behavior in multimodal pretraining. It also enables downstream applications such as action prediction for embodied AI. webpage+paper: bfl.ai/research/self-… code: github.com/black-forest-l… Credit to @hila_chefer, @pess_r, Dominik, @dustin_podell, Vikash, @Vinh_Suhi and Antonio. If you enjoy doing open research like this, come and join BFL! We are actively hiring🌲

New research from @bfl_ml 🥳 Meet Self-Flow: our self-supervised framework for image, audio, video & world models 🤖 bfl.ai/research/self-… Do generative models really need DINO to learn strong representations? We propose teaching them directly via a joint framework instead 🧵

@kilian_maciej @AnthropicAI Good stuff man :) Enjoy

excited to share that i'm joining the @AnthropicAI pretraining team! claude is by far my favorite model and it brings me so much joy to get to be part of this. everyone i've met here is brilliant and incredibly kind and i'm really excited to be working with them :)

@AnjneyMidha Beschleunigungsspur

if you were teaching a class to 500+ cs students on how to prepare for takeoff, what would you call it?

pencil autocomplete #3 realtime model: FLUX.2 [klein] by @bfl_ml via @fal

@steipete @sabinedoering Let's change it.

@sabinedoering Also, mindset. Wie ich nach 3 Jahren suchen wieder purpose gefunden hab: USA: "oh man this is so great let's build sth cool!" AT: "Ja aber pass schon auf di auf, net dassd wieder a Burnout kriegst, goi? Also mach a bissi langsamer."

Der OpenClaw-Gründer Peter Steinberger entkräftet hier sachlich viele B/Doomer-Bedrohungsszenarien. Nun geht er zu OpenAI. Warum konnte oder wollte Europa dieses Talent nicht halten? Ums Geld allein ging es ihm ja ganz offensichtlich nicht. on.orf.at/video/14311959…

@c_valenzuelab yes.

Within two years, 90% of the pixels you see on screen, from images and videos to games and software, will be generated by AI.

Flux 2 Klein 4B param, dropped at 2 steps with much higher FPS! I also added a couple of LORA's, I'll mess around with. Such a good diffusion model!

@venturetwins @jnack @krea_ai 👀

@jnack @krea_ai It’s powered by a new image model that’s much higher quality! But yes if you’re using professionally, you’d probably want to enhance ☺️

Real-time image editing is insanely good for architecture. You can take a sketch, render a photorealistic building, and then change the materials, weather, or environment by simply adjusting the prompt. Done with the new @krea_ai model in seconds 🤯

@DynamicWebPaige @nathanbenaich @airstreet @bfl_ml @TuscherMarc @SereactAI @stateofai Come visit!

@nathanbenaich @airstreet @robrombach @bfl_ml @TuscherMarc @SereactAI @stateofai fomo

i'm headed to zurich to host our @airstreet zurich ai meetup on 19 feb, ft: - @robrombach of @bfl_ml, frontier visual intelligence - @tuschermarc of @sereactai, general purpose robotics - an update on the @stateofai :) luma(dot)com/zurichai

Rocaille 2 vec2 p=(FC.xy*2.-r)/r.y/.3,v;for(float i,f;i++<1e1;o+=(cos(i+vec4(0,1,2,3))+1.)/6./length(v))for(v=p,f=0.;f++<9.;v+=sin(v.yx*f+i+t)/f);o=tanh(o*o);