Bria FIBO and its versions
Bria makes image models for companies that need to know where the training data came from: it says every FIBO model is trained only on licensed images. FIBO's other idea is the structured prompt, a JSON scene description the model writes from your text and you can edit. FIBO 2.0, which unifies generation and editing, is live on fal in October 2026 and has its own page with the same twelve prompts as every image model here.
Bria · facts checked
Which one to pick
Side by side, measured
| Model | Price | Seconds per image | 12 prompts cost | Sizes |
|---|---|---|---|---|
| Bria FIBO 2.0 | $0.04 per image via fal | 9 to 21 | $0.48 | About one megapixel at 21 named aspect ratios from 1:5 to 5:1, all four of ours included: 1024 square, 864 by 1152, 1152 by 864 and 1568 by 672 |
Seconds and cost are from our own runs of the same 12 prompts through each model; open a model for every picture and the request behind it.
Every version
| Version | Released | What changed |
|---|---|---|
| FIBO briaai/FIBO | Date not verified | The first FIBO: an 8 billion parameter text-to-image model driven by structured JSON prompts, trained on licensed data. Release date not verified. |
| FIBO 1.5 and FIBO Edit 1.5 briaai/Fibo-1.5 | Date not verified | Separate generation and edit models, before 2.0 merged them. Release date not verified. |
| FIBO 2.0 bria/fibo-2 | Date not verified | Generation and editing in one 8 billion parameter model in the FLUX.2 latent space, up to four images per edit, masks, and VGL structured prompts. $0.04 an image on fal. |
Frequently asked questions
Is Bria FIBO open source?
The FIBO 2.0 weights are on Hugging Face under CC BY-NC 4.0, so free for research and non-commercial use. Commercial use goes through Bria's API, a host such as fal, or an agreement with Bria.
What does trained on licensed data mean?
Bria says it trains only on images it has licensed, curated from a dataset of about a billion. It is the vendor's claim, and the reason enterprises choose it.