Happy to share something cool we made at @bria_ai_ - Zero-to-Wan: a minimalistic, easy-to-modify, finetuning repo for @Alibaba_Wan 1.3B video generation model - their most accessible one 📹 Check out the code here: https://t.co/42y9i3mpnp

🎬 Introducing 𝐁𝐫𝐢𝐚 𝐕𝐢𝐝𝐞𝐨 𝐄𝐫𝐚𝐬𝐞𝐫 Erase objects in 3 ways: ✍️ 𝑻𝒆𝒙𝒕-𝒃𝒂𝒔𝒆𝒅 (e.g. “remove the red car”) 🎯 𝑷𝒐𝒊𝒏𝒕 & 𝒄𝒍𝒊𝒄𝒌 auto-detection 🎭 𝑪𝒖𝒔𝒕𝒐𝒎 𝒎𝒂𝒔𝒌𝒔 for precise control Try it in the on @bria_ai_ lab: https://t.co/dsc8olaFWg Or on

🧵 New paper: "Simple Context Compression" - we show that mean-pooling beats the widely-used compression-tokens method for compressing contexts in LLMs, while being simpler and more efficient! with @yoavartzi (1/7)

If you missing publishing your research and contributing to the open community - my research group is hiring and provide competitive offering See our latest paper release in the comments Additional models and papers are already in the oven - great time to join us 🚀

🔥FIBO News: Paper & Training Code are out!🔥 Making the model fully open for the community to explore, reproduce and build upon is one of our core values. Eager to hear your thoughts 👉 Paper: https://t.co/KwuIKdbV8K 👉 Repo:

FIBO isn’t just another text-to-image model. It achieves SOTA benchmarks in aesthetics + alignment, is fast, and cost-effective. It generates structured JSON alongside each image, so the results are controllable and easy to integrate into agentic and enterprise workflows.

🚀 New open-source model drop: FIBO by @bria_ai_ FIBO transforms intent into structured JSON prompts for high-quality, repeatable image generation.

@bria_ai_ just dropped FIBO 🔥 ✨ 8B text-to-image model ✨ first open model to use json prompts natively! I tested it with one of the most complex prompts of all time and it did sooo well 🤯 >prompt: a man, a woman and a dog are standing against a backdrop, the backdrop is

Welcome 𝐅𝐈𝐁𝐎!! The world’s first JSON-native, Open Source T2I model! Winning both aesthetics and alignment in the benchmarks by being built ground up for 𝒄𝒐𝒏𝒕𝒓𝒐𝒍, 𝒑𝒓𝒆𝒄𝒊𝒔𝒊𝒐𝒏, and 𝒓𝒆𝒑𝒓𝒐𝒅𝒖𝒄𝒊𝒃𝒊𝒍𝒊𝒕𝒚 https://t.co/LMewUlpjzg The is big!

Fibo is a very cool model - you know exactly what it "sees" through a very detailed json prompt. You can easily refine and control your generations in an easy iterative flow 🎨

🚨 FIBO is now live on fal! 🧠 JSON-native text-to-image model for structured, controllable generation 🎛️ Edit lighting, camera, and color without breaking the scene 🧩 Trained exclusively on licensed datasets

Something big is coming.....

Incredible demo by @MokadyRon, announcing @bria_ai_'s new model, Fibo! The most controllable image model we've ever seen, and open-source 🚀

T2I models excel at realism, but true creativity means generating what doesn't exist yet. How do you prompt for something you can't describe? 🎨 We introduce VLM-Guided Adaptive Negative Prompting: inference time method that promotes creative image generation. 1/6

three years ago, DiT replaced the legacy unet with a transformer-based denoising backbone. we knew the bulky VAEs would be the next to go -- we just waited until we could do it right. today, we introduce Representation Autoencoders (RAE). >> Retire VAEs. Use RAEs. 👇(1/n)

I finally solved player recognition - player and number detection with RF-DETR - player tracking with SAM2 - team clustering with SigLIP, UMAP and KMeans - number recognition with SmolVLM2 stay tuned for YT tutorial: https://t.co/6Iq051w9Tn ↓ full breakdown + code

𝗡𝗲𝗲𝗱 𝘁𝗼 𝘂𝗽𝘀𝗰𝗮𝗹𝗲 𝗮 𝘃𝗶𝗱𝗲𝗼 𝘂𝗽 𝘁𝗼 𝟴𝗞? Bria Upscaler on fal = sharper, cleaner, flawless.

1/ What if you could animate a face directly from text? 🎭 Meet Express4D - a dataset of expressive 4D facial motions captured from natural language prompts, designed for generative models and animation pipelines. 🔗 https://t.co/okK0JgZMhV 📹👇

From my meetup talk a month ago: The biggest problem in text-to-image isn’t scaling or architectures. It’s evaluation. Today, we evaluate with “arenas”: user preference on short prompts. But humans struggle to follow long instructions, so to win you just optimize for average

Classic background removal uses binary masks, and that’s the problem. Real-world edges aren't just black or white. Soft alpha matting assigns per-pixel transparency and changes everything. Read my latest blog to see how we built it @bria_ai_:

Very excited about our ICCV work 🏄‍♂️ My favorite part: instead of using inversion, we just train the model to perform the identity function and used the emerged internal features at inference. Check the thread below for more details 😀