This is a 17B parameter self-merge of the 12B Flux.1-dev model, notable for being one of the first open-source 17B image models capable of generating coherent images. That's 5 billion extra parameters! FP8 quantized from fp16.

This is not an all-in-one (AIO) model, you still need t5xxl_fp8_e4m3fn_scaled.safetensors (or flant5xxl), clip_l.safetensors and flux_vae.safetensors.

Drag & drop any images I uploaded into ComfyUI as workflow is embedded!