Misha Feinstein

CTO @bria_ai_

[Misha Feinstein @MishaFein]

@bria_ai_ gets much more interesting in the hashtag #agentic era, what excites me here is the shift from capabilities to workflows.

https://t.co/KINKsCJJMy

This example turns #RMBG2_0 into something much more usable and creative:
- text-behind-image
- word cutouts
- layered effects
That’s the direction that matters: not just better models, but better ways to use them.
And the nice part is that background removal is only where it starts - the skill layer can unlock much more across the Bria stack.

[Misha Feinstein @MishaFein]

@bria_ai_ gets much more interesting in the #agentic era
What excites me here is the shift from capabilities to workflows.

https://t.co/QEawqfrLUM

This example turns #RMBG2_0 into something much more usable and creative:
- text-behind-image
- word cutouts
- layered effects

That’s the direction that matters: not just better models, but better ways to use them.
And the nice part is that background removal is only where it starts - the skill layer can unlock much more across the Bria stack.

[Saar Huberman @HubermanSaar]

Excited to share that our work, “Image Generation from Contextually Contradictory Prompts,” was accepted to CVPR 2026 🎉

Diffusion models fail on some seemingly simple prompts.
Why do they ignore what you asked for?
We show why, and how to fix it with a simple, training-free method.

Joint work with @OPatashnik @OmerDahary @MokadyRon @DanielCohenOr1

[BRIA AI @bria_ai_]

The Bria developer platform just got a meaningful overhaul!

The goal was simple: from the moment you discover an API to the moment you're confident integrating it into production, the experience should be fast, clear, and friction-free.

What's live now: the full Bria visual AI stack, all testable in sandbox before you write a line of code.

Generation. Editing. Video. Product shot. Background. Enhancement. Tailored models trained on your visual IP. Every endpoint. Every parameter visible. Every output reproducible. All powered by Bria’s Visual Generative Language.

If you're building products that touch visual content at scale, this is the fastest way to evaluate where Bria fits.

Free to test. No credit card. No sales call.

➡️https://t.co/tkjpTfAFdI

#VisualAI #GenerativeAI #DeveloperTools #APIFirst #Bria

[Ron Mokady @ CVPR @MokadyRon]

Making models run fast at inference requires optimizing the entire AI stack.

It was great partnering with MoonMath to take @bria_ai_ 's Fibo to the next level of speed. Unlike standard models, Fibo consists of a Reasoner (VLM) and a Renderer (Flow Matching), requiring both to be optimized at the algorithm, deployment, and kernel levels.

And most importantly it was great to work with @moonmathai

Read more in the new blog post

[Filip Strand @filipstrand]

MFLUX v0.17 is out with support for FIBO-Edit from Bria!

uv tool install --upgrade mflux

This model family includes state-of-the art editing capabilities with mask-support and also a model specifically tuned for quick and accurate background-removal.

https://t.co/KVWNZ6lIfJ

On this note, I'm also very happy to share that I've joined @bria_ai_ full-time!

#MLX #GenAI

[BRIA AI @bria_ai_]

Introducing Fibo, a next-generation AI image generation and refinement engine by BRIA.

Unlike traditional AI image tools, Fibo constructs visuals layer by layer, giving you full transparency into what changed and why. Simply describe what you want, refine with a prompt, and Fibo handles the rest — whether you're updating a product's color to match your brand, creating polished ad creatives, or iterating on visual concepts in real time.

✅ Image building for full creative control
✅ Intuitive prompt-based refinement — just tell it what to change
✅ Transparent AI: see exactly what was edited and why
✅ Built for ads, marketing, and brand-consistent visuals
✅ Deploy on BRIA's cloud — fast, scalable, and enterprise-ready

Try. Refine. Retry. - Now with Fibo, we finally can.

🔗Edit: https://t.co/NJWj0cj7UM
🔗Generate: https://t.co/LEx8uDetE6

[Misha Feinstein @MishaFein]

From the creators of #RMBG2_0 - we’re happy to introduce the open source #FIBO_Edit version for #BackgroundRemoval

Model card: https://t.co/JXsT33uazO

Why open source?
Because best use cases aren't invented in labs - they are done by the community.

Take the Model
Build your Solution.

What should @bria_ai_ open-source next?

[BRIA AI @bria_ai_]

Visual AI needs a language.

At #GTC2026 we’re introducing VGL — Visual Generation Language.

A structured way for developers, agents, and creative tools to generate and control images with precision.

If you’re building:
• AI agents that create visuals
• Programmatic image pipelines
• Brand-safe creative workflows

Watch the premiere here on X
Mar 17 • 9AM PST

[Misha Feinstein @MishaFein]

🔓 When we say open and 𝐜𝐨𝐧𝐭𝐫𝐨𝐥𝐥𝐞𝐝, we really mean it!

Today we’re releasing a research preview of our next‑gen models - with direct, disentangled control of visuals via both textual and 𝒏𝒖𝒎𝒆𝒓𝒊𝒄 𝒗𝒂𝒍𝒖𝒆𝒔 𝒊𝒏 𝒕𝒉𝒆 𝑽𝑮𝑳.

📄 Paper: https://t.co/uWcDkrSJ00
🧪 Demo in Bria AI lab: https://t.co/nSfo8KVWMs

[Ron Mokady @ CVPR @MokadyRon]

Research Preview:
Fibo BBQ: Bounding Box & Qolor Control in Large-Scale Text-to-Image Models

Text prompts are a terrible UI for precision
It’s much more intuitive to drag objects into place or use a color picker than to write “put it 20% left, make it teal, slightly bigger…”
Fibo BBQ demonstrates that we can train large-scale T2I models with numeric parameters (e.g., positions / boxes / colors) as part of a structured caption, at scale.

This is a step toward richer controllability: more “knobs” beyond text, and new UI paradigms that feel like design tools, not prompt engineering.

[Demo and Technical Paper below]

[Misha Feinstein @MishaFein]

Is that the best Creative Upscaler out there? 🤨

Probably yes! The new @bria_ai_ release just raised the bar.

Try on our platform: https://t.co/yHpaunHRuD
or
on @fal : https://t.co/7E4Avn92ie

[Replicate @replicate]

Video Background Removal from @bria_ai_ is now on Replicate. It’s consistently SOTA & built for real production workflows 👇

[BRIA AI @bria_ai_]

Try our Video Background Removal on @replicate.

#YourAIYourRules

[Gal Davidi @gal_davidi1]

just shipped an AI photoshop for your agent.
it's called Bria AI Skills.
15+ tools. 13+ edit modes. full pipeline workflows. structured JSON prompts designed for precision, not vibes.

[Misha Feinstein @MishaFein]

Welcome 𝐅𝐢𝐛𝐨-𝐄𝐝𝐢𝐭 to the 𝐅𝐢𝐛𝐨 family 👋

Bria AI’s new foundation model for deterministic image editing.

🧩 Fibo builds the image. Fibo-Edit perfects it - using structured JSON control for pixel-perfect edits.

Highlights:
⚡ 8B params, production-ready
🎯 Native masking & parametric edits (VGL)
🤗 Open-source on @huggingface + APIs
🛡️ 100% licensed data, safe for commercial use
Get started:
• Weights: https://t.co/N8GTSnfjyE
• Demo: https://t.co/J22LTcCrg6
• Run on @fal : https://t.co/4qGFvzq2q5

Who’s next in the Fibo family? 👇

#FiboEdit #BriaAI #OpenSource #GenAI #ResponsibleAI

[Misha Feinstein @MishaFein]

Language gives you control. Control makes AI yours.
Stop guessing with prompts. Start building with parameters. Read the full documentation on the VGL paradigm shift here: 👉 https://t.co/Bx1E7fWk59 [8/8]

#GenerativeAI #VGL #BriaAI #DevTools #ParadigmShift

[Misha Feinstein @MishaFein]

Generative AI has a control problem. 🎲 You prompt, you pray, and you get a "Black Box" result. If it’s wrong, you re-roll the dice.

For enterprise and production, "Prompt Engineering" isn't enough. It’s time for a paradigm shift: Visual GenAI Language (VGL). 🧵 [1/8]👇

[Misha Feinstein @MishaFein]

The Schema. 📝 VGL breaks an image down into controllable nodes: • objects: bounding boxes, poses, clothing. • lighting: direction, shadows, conditions. • camera: Angles, lenses. • aesthetics: Mood, style, palette.
You control the source code of the image [7/8]