What FIBO 2.0 Is
According to the Fibo 2.0 Dev model card, FIBO 2.0 is Bria AI’s next-generation unified model for image editing and text-to-image generation. It is an 8B-parameter flow-matching model operating in the latent space of the FLUX.2 VAE, using a single-stream DiT with 44 blocks and prompts encoded by Qwen3-VL-4B-Instruct. Bria trained the model exclusively on licensed data using structured JSON captions (Visual GenAI Language, or VGL) at roughly one megapixel (such as 1024x1024 default). Its exact release date is not verified.
The full non-distilled base model, Fibo 2.0 Dev, is designed for fine-tuning and runs with classifier-free guidance (50 steps and guidance 5.0 recommended). It supports five modes: edit, masked edit, multi-reference (up to 4 reference images), generate, and inspire. Bria also offers Fibo 2.0 Edit Turbo, a distilled editing-only variant that completes edits in 4 steps without classifier-free guidance. On its model card, Bria states that FIBO 2.0 ranked #1 in image editing and #2 in text-to-image among open-weights models on Artificial Analysis as of October 20, 2026. You can also access the model through Bria’s API, though the API price per image is not verified.
License: CC BY-NC 4.0 Means No Commercial Use of the Weights
Both Fibo 2.0 Dev and Fibo 2.0 Edit Turbo are gated on Hugging Face under the CC BY-NC 4.0 license. In plain terms, you can download and use the open weights for research, testing, and personal non-commercial projects, but the license does not allow commercial use of the weights.
To use FIBO 2.0 in a business setting, you must sign a separate commercial agreement with Bria. As noted in Bria’s self-hosting documentation, running the model on your own servers commercially requires a paid self-hosted license.
By contrast, FLUX 3 Klein uses the FLUX Kommunity License v1.0, which permits non-commercial use plus commercial output generation by qualifying users under US$5M in annual revenue (subject to license terms).
FIBO 2.0 vs FLUX 3 Klein Comparison
| Feature | FIBO 2.0 (Dev) | FLUX 3 Klein |
|---|---|---|
| Parameters | 8B (source) | 5.6B (source) |
| Text Encoder | Qwen3-VL-4B-Instruct | Qwen3-VL-4B (source) |
| Reference Images | Up to 4 | Up to 10 |
| Max Output | Trained at ~1 MP (1024x1024 default); max not verified | Up to 4K |
| License | CC BY-NC 4.0 (commercial requires paid Bria agreement) | FLUX Kommunity License v1.0 (commercial outputs allowed under US$5M revenue, subject to terms) |
| Local Hardware | NVIDIA GPU, ~26 GB weights (source); minimum VRAM not verified | Fits consumer GPUs down to RTX 3090 (source) |
| API Pricing | Available via Bria API; price per image not verified | $0.009 per 1K image (source) |
| Try It on Wispvue | No (not supported) | Yes |
Which Model Should You Choose?
Choose FIBO 2.0 if your priority is training-data provenance (trained exclusively on licensed data), structured JSON prompting via VGL, or fast 4-step edits using Edit Turbo—provided your work is strictly non-commercial or you are ready to purchase a commercial license from Bria.
Choose FLUX 3 Klein if you want a smaller 5.6B model that fits consumer GPUs down to an RTX 3090, supports up to 10 reference images, outputs up to 4K resolution, and allows commercial output generation for qualifying users under US$5M in revenue.
Note: Wispvue (flux3klein.org) is an independent, unofficial website not affiliated with Black Forest Labs or Bria AI. Wispvue does not run or offer FIBO 2.0 generation; our platform only runs FLUX 3 Klein.