Based on https://github.com/rickrender/FluxLatentDetailer/
This workflow uses araminta_k_flux_koda.safetensors which can be found at CivitAI.
https://civitai.com/models/653093/Koda%20Diffusion%20(Flux)
As well as PARP:FLUX - https://civitai.com/models/679120/parpflux?modelVersionId=760166
And mstc3:flux-v2 - https://civitai.com/models/682052/mst3cflux-v2?modelVersionId=763389
It also employs a slightly improved CLIP-L model, available from here: https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/blob/main/ViT-L-14-BEST-smooth-GmP-TE-only-HF-format.safetensors place this into ComfyUI\models\clip.
The Kodachrome LUT file can be downloaded from here: https://www.freepresets.com/product/free-lut-kodachrome-lookup-table/
LUT goes into ComfyUI\custom_nodes\ComfyUI_essentials\luts, if you are using cubiq's comfy_essential nodes like in this workflow.
Changes:
GGUF loader for Quantized models, Prompts Everywhere, Anything Everywhere, negative prompt for first sampler (if needed), inpainting support and IMG2IMG. Also added an improved CLIP-L model, linked above, for better prompt adherence.
Note: Inpainting, when tested at 1.00 denoise, caused the whole image to get really crummy in the 2nd and 3rd step. I believe this is a limitation of how I setup inpainting thusfar, and will be fixed in the future.
-immoralhole0