Fibo 2.0 is Bria's new open-weights image model for marketing, agency, and product teams that need campaign visuals they can revise without starting over. The 8-billion-parameter model generates from a short prompt and edits from an instruction, including multi-reference edits that combine up to four images, so a product, a setting, and a wordmark can land in one frame while identity, layout, and lighting you did not ask to change stay put. Each result comes back with a structured JSON scene description, Visual GenAI Language (VGL). Change lighting, camera, color, or one object in that JSON, keep the seed, and the next render follows the edit. Artificial Analysis ranks Fibo 2.0 first among open-weights models on its image editing leaderboard. It is statistically tied with Qwen-Image-2.1 for the open-weights lead on text-to-image, and it scores higher than Bria's earlier Fibo models on every editing action and text-to-image capability they measure, with the largest gains in text edits, text rendering, and layout. Bria reports that it also edits better than closed models from Ideogram, ByteDance, and Black Forest Labs. The weights are on Hugging Face as Fibo 2.0 Dev, for generation and fine-tuning, and a four-step Edit Turbo for faster edits, under a non-commercial CC BY-NC 4.0 license. Commercial use needs a separate agreement with Bria, the Tel Aviv company founded in 2020 by Dr. Yair Adato. Training uses only licensed data, the API is priced at $32 per 1,000 images, and existing generate and edit calls keep the same endpoints. The model is already live on Bria's platform and on fal.