AI BriefingFibo-2.0: Fibo 2.0: Image Generation with 1,000-Character Captions
Comfy-Org/Fibo-2.0
About the project
This file repackages BRIA AI's Fibo 2.0 model for immediate use in the ComfyUI environment. It consolidates the distributed weights from the original repository into a single file structure recognized by ComfyUI. It includes Qwen3VL 4B as the text encoder and Flux2 as the VAE.
It supports both the original bf16 precision weights and an int8 quantized version. The int8 version applies Hadamard rotation and per-channel scales to improve memory efficiency, while keeping only the adaLN modulation layers in bf16. Output results may differ slightly from the original due to the quantization process.
This model converts text to images using structured captions proposed in the 'Generating an Image From 1,000 Words' paper. It is characterized by accepting detailed descriptions rather than simple prompts to derive more precise generation results.
The license is CC BY-NC 4.0, allowing only non-commercial use. For commercial use, a separate license agreement must be signed with BRIA AI. It is suitable for developers using ComfyUI who want to build text-based image generation pipelines.
Comfy-Org/Fibo-2.0
The original page has no description.
image-to-image
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.