A simple Chroma workflow that allows GGUF models with one LoRa for fast generations. Silveroxides quantized models can be found at https://huggingface.co/silveroxides/Chroma-GGUF/tree/main and accelerating LoRAs at https://huggingface.co/silveroxides/Chroma-LoRA-Experiments/tree/main. Lodestones amazing original work here: https://huggingface.co/lodestones/Chroma/tree/main

V 1.1

Adapted my simple GGUF & LoRA WF so it can work with Nunchaku (needs Comfy_Nunchaku v0.3.3 or 0.3.4 to be able to load a folder. Higher versions no longer support that feature).

1. Download from https://huggingface.co/rocca/chroma-nunchaku-test/tree/main/v38-detail-calibrated-32steps-cfg4.5-1024px the files: comfy_config.json, config.json, transformer_blocks.safetensors and unquantized_layers.safetensors and put them in a folder, say Chroma_Nun in the diffusion_models folder of ComfyUI.

2. In the workflow, use the Nunchaku_Flux_DIT_Loader node to select the above folder/path.

3. Get creative and enjoy the speed.

A big thanks to Joe Rocca for quantizing Chroma v38 for Nunchaku

Edit: The first workflow I uploaded had the wrong LoRA nodes. It should be fixed. Re-download.

V 1.2

Added an Ultimate SD Upscale node for higher resolutions. Recommended sampler/scheduler: res_2s/bong_tangent for quality, euler/simple for speed, depending on the situation or subject.