https://www.liblib.art/modelinfo/0dd2b45a7ff5467fbbb9c5d8eb50f2e8

Chroma is a large 8.9 billion parameter text-to-image generation model maintained by community developer lodestones and optimized based on the FLUX.1-schnell architecture, using the rectifier stream Transformer architecture, with high-detail restoration, no built-in censorship, and ComfyUI ecological adaptation as its core advantages, suitable for high-precision creative generation and model secondary development. The training focuses on strengthening the restoration of human structure, hand and facial details, and solves the common proportion imbalance problem of traditional models. Suitable for medical illustrations that require precise anatomical representation.

5 million selected samples of training data (from 20 million candidate sets), including animation, realism, art, photography, etc., marked with the aesthetic label "aesthetic0 -11".

Sampling and guidance mechanism

It is compatible with mainstream samplers such as Heun, DPM++, and Eulera