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 š
Description
The initial release is as simple as possible to help you start learning and playing with it.
FAQ
Comments (19)
Workflow isn't complete it errors without any modification. Annoying they don't test WF if they are selling early access. You have to add VAE model and decoder. Workflow is basic and is more about how use advanced samplers of RES4LYF.
Really simple WF and it works perfectly fine here...
Workflow worked for me with no issues at all. Try troubleshooting your errors before dissing someone else's work.
oh, don't tell anyone/everyone - that you can't load VAE (which is soooo simple...), and by the way - workflow works like a charm out of the box (if you are familiar with comfyui)
and btw, nobody was forcing to pay anybody for anything - since I was uploading as samples PNGs with full metadata / comfyui workflow inside completely for free ... you know, for smart people ;)
OliviaRossiĀ you are a real class act.
ckĀ š( ā¢_ā¢ ) š
The WorkFlow works like it should out of the box, creates amazing outputs, but you have to install all missing nodes first! (if any)
and TY everyone who BUZZED the WF <3
Apologies if I'm misunderstanding but this is just RES4LYF's Chroma example workflow, you just swapped out the Clown loader for GGUF. Am I missing something here? https://github.com/ClownsharkBatwing/RES4LYF/blob/main/example_workflows/chroma%20txt2img%20WF.jpg
well, the initial release was on purpose as simple to understand as possible, and in case of Chroma we can do more than just swapping with ClownShark samplers. In this version we simply have to remove Padding (it is not compatible) ... and those samplers are way better at realism / art generation with Flow-type models (Flux ... etc.) - the only thing is that we have to find the best one sampler/scheduler combo for our desired output ourselves experimentally. And regular KSampler is not listing those "additional" samplers properly - so it is better to use it (the ClownShark one). Also, instead of the default text encoder - the FLAN Text-Only is being used.
The very next version of the workflow will be more complex than this one ;)
OliviaRossiĀ i do miss my xyz scripts on comfyui, olivio should make a tutorial :)))
MetaGenĀ agree
Workflow work perfectly, but i can't have same image ,why ?
it depends on the text encoder - what kind of it you are using (as you can see in my WF - it is a FLAN Q8 Text-Only), also it heavily depends on the sampler / scheduler combo - one of them is better for art while the other one - for photorealism
OliviaRossiĀ I Used same, from your WF, but not same result
ghobi12 for testing and debugging - you can download one of my sample images as PNG file (with metadata) and drag&drop them on the ComfyUI, also - I'm using Chroma 48 - detail corrected Q8, the image size has to be the same, ... also I'm running out of ideas why you are getting different image, maybe your Comfyui version is different - the Pytorch library is not the very last one ... idk
OliviaRossiĀ Humm Ok, Thanks
ghobi12Ā I hope you forced it to do what you want
OliviaRossiĀ What's the difference with the Flan text encoder vs the the normal t5xxl?
OnlyGensĀ FLAN is more superior to XXL, also we are using TE = only the text encoder part - less memory consumption. But it means nothing, in fact you simply have to run once the same generation with both and decide for yourself which one is better for your goals.