I was not satisfied with the existing Chroma workflows for their ability to reliably generate high-quality images. After some research, I created this workflow for myself!
You will need to visit the following resources for the models you need:
1) Chroma GGUF: https://huggingface.co/silveroxides/Chroma-GGUF/tree/main
2) CLIP GGUF: https://huggingface.co/silveroxides/flan-t5-xxl-encoder-only-GGUF/tree/main
3) upscale model: https://huggingface.co/Kim2091/UltraSharpV2/tree/main
Also, (most importantly!), you absolutely must familiarize yourself with the following project: https://github.com/ClownsharkBatwing/RES4LYF <-- to understand WTF is res_3m / beta57 or res_2s / bong_tangent ;)
Feel free to experiment with the sampler/scheduler combinations, as they have a significant impact on the output!
P.S.: use it well 😏
Description
This version is more complex: while we avoided excessive overcomplication, we pushed harder for an even higher quality of image generation. You will find some advanced methods here. ;)
FAQ
Comments (4)
Idk why would you use the old latent upscaler with 4 sampler nodes in the case
that is is how those nodes are working - through the latent upscale there is an information transfer to the next nodes (needed by the architecture of those cyclical samplers), and after the initial latent upscale - first there is a unsampling and then resampling = all of it was proposed by the author of those nodes as more high quality method of denoising - those are not regular KSamplers!
as it is mentioned in description of the WF: get familiar with the way those custom KSamplers work to understand the method and math behind them
There are some good quality improving ideas, though no latent upscale anywhere.
It's image upscale. There is however upscale latent with VAE which is not present here, and does other things entirely.
I love a lot of this for a single tech-demo maximalist approach, and it does some things better but just as much worse than a plain Ultimate SD Upscale... but it does so in 330 seconds versus 31 seconds.
@yorgash could you make a workflow that swaps that in for this please?