Bria FIBO 2.0 in ComfyUI: Text to Image and Image Edit

FIBO 2.0 is Bria’s 8B image model that generates and edits images with one set of weights, and it runs in ComfyUI as Partner Nodes. It keeps the structured-caption training the FIBO line was built on: prompts are expanded into a Visual GenAI Language (VGL) caption that names lighting, composition, color, camera, and depth of field as separate fields, and both nodes return that caption as a structured_prompt JSON string. That JSON is what makes FIBO predictable. Feed it back with the same seed to rebuild an image, change one field to refine it, or build it programmatically. FIBO 2.0 adds instruction editing on the same model: describe the change in plain language, or hand over the structured prompt, and everything you did not ask to change stays put. Both nodes are cloud API nodes. They need a Comfy account with credits, download no model, and bill per run; see Partner Node pricing for the rates. This page also covers the Bria outpainting template, which runs on the Bria Expand Image node.

​

What Bria FIBO 2.0 is good at

• One model for both directions: the same FIBO 2.0 weights generate a new image or edit an existing one
• Structured control: lighting, composition, camera, and style are explicit JSON fields instead of prompt guesswork
• Edits that hold: identity, layout, lighting, and fine detail stay intact outside the area you asked to change
• Typography and texture: legible in-image text, convincing material surfaces, and more reliable anatomy than earlier FIBO releases
• Licensed data: trained on licensed data, with the provenance commercial work needs

FIBO 2.0 handles the edits teams ask for most:

• Time of day and relighting: change the hour or the light source without rebuilding the scene
• Material and texture swaps: fabric, metal, finish, and clothing
• Masked area editing: confine a change to a painted region and keep the rest pixel-identical
• Text adjustment: rewrite or restyle the words inside an image
• Restyling: re-render a photo in a different art direction
• Posters and covers: text-led layouts where the headline has to be rendered literally

​

Example outputs

An instruction edit on the FIBO 2.0 Image Edit workflow, keeping the rooftop scene and changing only the light:

​

Bria: FIBO 2.0 Text to Image

Describe the image, pick a shape, and run. FIBO 2.0 expands the prompt into a structured caption, generates at 1 MP or 4 MP, and returns the caption next to the image so you can rebuild or refine it later. How to use

1. Write the prompt: describe the image in prompt. FIBO 2.0 follows long, specific briefs closely, including the exact wording that should appear inside the image.
2. Pick the shape: aspect_ratio defaults to 1:1, and the presets cover 16:9, 9:16, 4:3, 3:2 and the wider 2:1, 21:9, 3:1, 5:1 formats.
3. Pick the resolution: resolution is 1MP or 4MP.
4. Optional reference: connect a photo to image to steer style or content. The result is a new composition at the selected ratio, not an edit of that photo, and it cannot be combined with structured_prompt.
5. Run and iterate: seed reproduces a result. The structured_prompt output holds the JSON caption, so you can feed it into another FIBO 2.0 node to refine a single field or rebuild the same image later.

​

Bria: FIBO 2.0 Image Edit

Edit an image by describing the change, with an optional mask to confine it to one region. FIBO 2.0 holds identity, layout, and detail outside the edited area, and returns the structured prompt for the edit so it can be rebuilt or refined from JSON. Input materials Upload this file to the matching LoadImage node: How to use

1. Load the image: upload it in LoadImage and connect it to BriaImageEditNode.
2. Describe the edit: write the change in prompt, such as Modern architectural photography at night, blue hour. Use negative_prompt to keep unwanted elements out.
3. Or use JSON: paste a structured_prompt from an earlier FIBO 2.0 result for precise, repeatable control. The node takes one of prompt or structured_prompt, not both.
4. Optional mask: paint the region to regenerate and connect it to mask. With no mask the edit applies to the whole image, and a mask that does not match the image size is rejected.
5. Tune and run: guidance_scale runs from 3 to 5 (higher follows the prompt more closely) and steps from 20 to 50. seed reproduces a result.

​

Bria: Image Outpainting

A professional Bria-powered outpainting template designed for seamless image extension and edge-aware scene expansion. Input materials Upload this file to the matching LoadImage node: