A beginner's guide to the Fibo model by Bria on Replicate

#coding #ai #machinelearning #programming

This is a simplified guide to an AI model called Fibo maintained by Bria. If you like these kinds of analysis, you should join AImodels.fyi or follow us on Twitter.

Model overview

fibo is a state-of-the-art open-source text-to-image model built for enterprise and agentic workflows. With 8 billion parameters trained exclusively on licensed data, it prioritizes controllability and precision over casual image generation. Unlike most text-to-image models that excel at imagination but lack control, fibo uses structured JSON captions up to 1,000+ words to enable reproducible control over lighting, composition, color, and camera settings. This training approach on long structured captions fosters native disentanglement, allowing targeted refinement without prompt drift. Compared to alternatives like image-3.2, which focuses on aesthetic quality with 4 billion parameters, fibo trades parameter count for structural precision and iterative control capabilities.

Model inputs and outputs

fibo accepts flexible inputs depending on the generation mode and produces high-quality images with accompanying structured prompts. The model can work with text prompts alone, detailed structured JSON prompts from previous generations, reference images, or combinations of these inputs. The output includes both the generated image and an expanded or refined structured prompt in JSON format, enabling reproducibility and further iteration.

Inputs

Prompt: Text description of the desired image
Structured prompt: Detailed JSON-formatted prompt from a previous generation (optional)
Image: Reference image for inspiration or refinement (optional)
Seed: Integer for reproducible generation
Negative prompt: Text describing what to exclude from generation
Guidance scale: Numerical value (3-5) controlling prompt adherence strength
Aspect ratio: Output dimensions (1:1, 16:9, 9:16, and others)