AI Briefingflux-3-image-klein-comfy: Control image layout with coordinate boxes in prompts
black-forest-labs/flux-3-image-klein-comfy
About the project
Precisely adjust the position and size of image elements by including coordinate boxes ranging from 0 to 1000 within text instructions. It supports text rendering in generated images and handles tasks from single edits to multi-image compositing with a single model, accepting up to 10 reference images.
The architecture combines a 5.6B parameter image transformer with the Qwen3-VL-4B-Instruct text encoder. Various precision checkpoints, including BF16, int8, and NVFP4, are provided for the ComfyUI environment. The int8 version runs on 8GB VRAM GPUs such as the RTX 3070, lowering hardware barriers.
The distilled model enables fast generation in 16 steps, while the base model allows fine-tuning with 50 steps and CFG 3.5. A prompt upsampler is included to convert short requests into detailed scene captions, reducing the burden of prompt writing. Under the FLUX Kommunity license, non-commercial use and commercial output generation below a certain revenue threshold are permitted.
black-forest-labs/flux-3-image-klein-comfy
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.