FIBO - Generate Image

If you only install one node from the BRIA AI API nodes pack, this is probably it. FIBO - Generate Image is the flagship: text-to-image, image-to-image, or "recreate exactly" from a saved structured prompt, all through one node. The heavy lifting happens on BRIA's servers, the image comes back as a normal ComfyUI IMAGE tensor, and the only local prerequisite is an API token from platform.bria.ai.

The "FIBO" in the name is the model - BRIA's 8B-parameter generator trained on structured JSON captions up to 1,000+ words, built for professional control rather than casual novelty. It generates through a two-step pipeline: first a VLM bridge (this node uses Gemini 2.5 Flash - the Lite sibling uses BRIA's open-source FIBO-VLM) translates your text into a machine-readable structured prompt, then FIBO renders from that JSON. Community reception was warm: "works nicely with JSON structured prompts and with really good prompt adherence", "the least slopped new image model of 2025". Hands still trip it up occasionally; prompt adherence is its calling card.

The inputs that matter

• api_token and prompt - required. The placeholder BRIA_API_TOKEN is rejected until you paste a real key.
• aspect_ratio - 1:1 through 16:9, default 1:1.
• steps_num - 35–50, default 50. The full node's window is higher than Lite's because Gemini does more detailed translation; keep 50 for quality.
• guidance_scale - 3–5, default 5. FIBO's whole range - don't go hunting for SDXL-style 7s.
• seed - a STRING, accepting comma-separated values for multi-image runs. Same seed + same structured prompt = pixel-repeatable.
• negative_prompt - optional (the one thing Lite lacks).
• images - feed reference image(s) for image-conditioned generation. The node handles batches, running one API call per image.
• structured_prompt - the reproducibility lever. Paste a JSON from a previous run (the node outputs it) and you can recreate the exact image with the same seed.

Outputs: image, plus the structured_prompt JSON that was actually used and the seed - wire both into text/Show nodes if you want to build a repeatable library.

Install

No requirements file, no model downloads - the pack only uses libraries ComfyUI already ships:

cd path_to_comfyui/custom_nodes
git clone https://github.com/Bria-AI/ComfyUI-BRIA-API.git

or search "BRIA API" in ComfyUI Manager, restart, and drop in your token.

How to not burn money

Every run is a paid API call, and the worst way to use this node is to iterate like you're at a local KSampler. Lock a seed, write your prompt the way FIBO likes it - structured, specific about camera and lighting, no masterpiece, best quality boilerplate, which is dead vocabulary on LLM-encoded models like this - and only then spend generations. If you want to see the JSON translation before committing, run FIBO - Generate Structured Prompt first and audit the output; that's the human-in-the-loop workflow this pack was designed around. There are ready-made generate_image_fibo_workflow.json and lite variants in the pack's workflows folder.