FIBO - Generate Image - Lite

This is the "cheaper twin" of BRIA's flagship FIBO generation node. FIBO - Generate Image - Lite runs the same two-step pipeline - a VLM translates your text into a structured JSON prompt, then the FIBO model renders the image from that JSON - but the translation step uses FIBO-VLM, BRIA's own open-source bridge, instead of Gemini 2.5 Flash. The result: faster and cheaper generation, at the cost of slightly less clever prompt translation. The full node gives you the most detailed structured prompts; Lite gives you speed and lower cost per image.

Both nodes are API clients in the official BRIA AI API nodes pack - everything happens on BRIA's servers, your GPU sits idle, and you need an API token from platform.bria.ai. FIBO itself is the 8B-parameter model BRIA trained on structured JSON captions up to 1,000+ words, released in late 2025 to a community that largely liked it - "really good prompt adherence", "least slopped new image model of 2025", with hands still a known weakness.

How it works

You type a prompt, the node POSTs it to BRIA's /v2/image/generate/lite endpoint, the FIBO-VLM bridge turns it into a structured prompt, FIBO renders, and the node polls the status URL until the image is ready. It downloads the result and hands you back a ComfyUI IMAGE tensor. The structured prompt that was actually used comes back as an output too - useful for reproducing or reusing the translation later.

The inputs that matter

• api_token and prompt - required. Swap the BRIA_API_TOKEN placeholder for your real key before your first run.
• aspect_ratio - nine choices from 1:1 to 16:9, default 1:1.
• steps_num - 8–30, default 8. Lite's ceiling is low (30) because it's optimized for speed; the full node goes up to 50.
• guidance_scale - 3–5, default 5. This is FIBO's whole CFG range; it's not SDXL, don't hunt for a 7.
• seed - a STRING (it accepts comma-separated values for batches). Same seed + same structured prompt = same image.
• images - optional reference images for image-conditioned generation; structured_prompt lets you feed in a pre-made JSON and skip translation.

Outputs: image, plus structured_prompt and seed strings that feed the next node or log your reproducibility.

Install

No requirements, no model downloads:

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

or ComfyUI Manager → search "BRIA API" → restart, and paste your token.

Lite vs. full - which one

Honestly, start with Lite. For most prompts you won't see a meaningful quality difference, and you'll burn fewer credits while you learn how FIBO wants to be prompted (structured, block-y, specific about camera and lighting - the 2026 instruction-style rules from the KB's prompt-engineering essay apply). Move up to the full node when you need the richer Gemini translation for complex scenes or heavy multi-image conditioning. And if you want to see the JSON before it costs a generation, chain FIBO - Generate Structured Prompt - Lite into a text viewer instead.