Prompting uses natural language as the interface to high-dimensional visual models. Natural language is expressive, but fundamentally imprecise.

The structural limitations:

• Cannot specify exact spatial layout

• Describes attributes qualitatively, not quantitatively

• Multiple distinct images satisfy the same prompt

• Small changes produce large, unpredictable effects

Bria introduces a formal control language for generative visual systems.

Rather than words alone, this language encodes visual intent through:

• Geometry- Spatial coordinates, bounding boxes, depth maps)

• Color- Deterministic RGB/HSL values, palettes

• Camera- Angle, FOV, focal length, distance

• Lighting- Direction, intensity, color temperature

• Composition- Object relationships, layering, focus

• Attributes- Material, texture, style parameters

Fibo is Bria's foundation model family, trained natively on structured visual representations.

Instead of inferring intent from sparse text, Fibo consumes explicit visual state descriptions:

• Deterministic generation- Same input, same output. Every time.

• Inspectable state- See exactly what parameters created any image

• Programmatic modification- Change specific values without regenerating everything)

• Automation-ready- Structured inputs that agents and pipelines can manipulate)

• Move an object- Update position coordinates. Everything else locked.

• Change lighting- Modify light parameters. Composition unchanged.

• Swap backgrounds- Replace scene values. Subject stays pixel-perfect.

How does Fibo work?