BRIA launches Fibo engine
BRIA AI unveiled the Fibo engine for layer-by-layer image generation and refinement aimed at brand-consistent creatives and color-matching tasks, touting transparent edits and cloud deployment for studio pipelines. The demo emphasizes control and repeatability—features that make AI-generated visual assets easier to integrate into post-production chains. (x.com)
Bria.ai announced FIBO as a new deterministic visual foundation model on Nov. 3, 2025 in a company release positioning it for enterprise production workflows. (prnewswire.com) The project’s public repository describes FIBO as a JSON‑native model trained on structured captions up to 1,000+ words to enable fine‑grained, attribute-level prompts. (github.com) The GitHub tree shows active development including a “Fibo Lite” branch and an explicit LICENSE file in the repo history, indicating Bria is publishing source code and weights alongside documentation and a forthcoming technical report. (github.com) Bria states FIBO was trained on 100% licensed, multi‑domain data with provenance tracking, C2PA watermarking, and claims readiness for EU AI Act compliance to support enterprise legal and governance requirements. (pages.bria.ai) Bria exposes FIBO through its platform API and lists Fal.ai and Replicate as hosted endpoints for cloud access, while the model is also packaged for local use via ComfyUI partner nodes for studio integration. (huggingface.co) (docs.comfy.org) A Docker‑ready “Fibo Lite” variant is published for on‑premise, lower‑latency inference aimed at studios that require local deployment rather than cloud-hosted inference. (pages.bria.ai) Bria distributes FIBO under an open‑source, non‑commercial license on public model hubs and states commercial use requires a Bria commercial license and enterprise terms for indemnification and production SLAs. (huggingface.co) (fiboai.org)
