FIBO Edit: Structured, Deterministic Image Editing

Text-to-image models have mastered creation. FIBO Edit masters control.

FIBO Edit is the first JSON-native image editing model. Unlike traditional editors that rely on vague text prompts (“make it look vintage”), FIBO Edit operates on structured data. It translates your instructions into explicit JSON blueprints controlling lighting, composition, style, and camera settings.

With just 8 billion parameters, it delivers production-grade fidelity, zero “prompt drift,” and fully disentangled control. Built on 100% licensed data, it is the safe, repeatable choice for enterprise pipelines.

🌍 Why FIBO Edit?

Standard generative editing is a guessing game. You ask for a “sunset,” and the model accidentally changes the subject’s face or the background texture.

FIBO Edit changes that. By using Visual GenAI Language (VGL), it decouples the edit instruction from the image content. It updates only the specific parameters you target—whether that’s changing a camera lens to 85mm, shifting the lighting to “Golden Hour,” or swapping a background—while mathematically preserving the rest of your scene.

🔑 Key Features

• JSON-Native Control: Edits are defined by structured schemas, not just text. Control lighting, depth of field, and composition explicitly.
• Precision Masking: Native support for mask-based editing allows you to target pixel-perfect regions while freezing the rest of the image.
• Disentangled Editing: Tweak a single attribute (e.g., camera_angle: low_angle) without breaking the scene’s consistency.
• Enterprise-Grade: Trained exclusively on licensed, rights-cleared data. Safe for commercial production.
• VGL Paradigm: Uses a Vision-Language Model (VLM) bridge to translate human instructions into machine-perfect execution.

💡 What can you do with FIBO Edit?

1. Global Edit

Give a natural language instruction like “change the lighting to neon cyberpunk.” FIBO Edit translates this into a structured JSON layer, applying the style transformation globally while keeping the original structure intact.

2. Targeted Masking

Upload an image and a mask. FIBO Edit respects the boundary perfectly, regenerating only the masked area (e.g., “replace the coffee cup with a vintage teacup”) while ensuring lighting and shadows blend seamlessly with the unmasked original.

3. Iterative Refinement

For professional workflows, you don’t have to start from scratch. Pass a structured_instruction JSON from a previous run to lock in your settings, then tweak just one value (e.g., change "focus": "soft" to "focus": "sharp") for a deterministic, non-destructive update.

🚀 Usage

Input Parameters

• image: (Required) The source image to edit.
• instruction: (Required) Text description of the desired edit (e.g., “make it look like a pencil sketch”).
• mask: (Optional) Black and white mask image. White areas will be edited; black areas preserved.
• structured_instruction: (Optional) Bypass the VLM and pass raw JSON for precise, programmatic control.

Source-Code & Weights

• Non-Commercial Use: The model is open source for non-commercial use with this license: Creative Commons Attribution-NonCommercial 4.0 International.
• Commercial Use: Click here for commercial licensing.