a pile of junk-MixMod-workflow - CivArchive (CivitAI Archive)

■This workflow uses 18GB of VRAM if not optimized.

And honestly, there's a 99% chance it's not worth trying for you.

It's a self-satisfying setup, so I’d appreciate it if you just see it as someone doing strange experiments.

■This is a workflow built around an extension called MixMod.

This extension allows you to combine multiple models that use the same VAE.

For more details, please refer to the repository or discussion thread—it should be easier to understand there.

Although I’m using it in a rather unusual way, MixMod itself is a casual and fun feature—you can combine your favorite models and enjoy experimenting with them. I definitely recommend giving it a try!

As a result, I’ve been able to combine PixArt-Sigma and SDXL more deeply, since they use the same VAE.

■And this workflow takes it even further with a very specific use case.

It combines my fine-tuned PixArt-Sigma 512px and Playground-v2 512px, and then refines the result further through i2i upscaling with SD1.5.

The concept behind this workflow is the reuse of beloved "junk"—

Playground-v2 builds a unique style, PixArt-Sigma strengthens conceptual understanding, and SD1.5 polishes the final details.

Each model compensates for the weaknesses of the others while enhancing their strengths.

■However, even after all this effort, it doesn’t produce results that surpass other models—so please don’t expect too much.Also, instead of forcing all three models together, separating the workflow—like Playground + SD1.5 or PixArt + SD1.5—can reduce VRAM usage and better preserve each model's unique strengths. I think that approach might be more effective.

■Below is a list of what you'll likely need for the workflow to function properly.

If you're using an old ComfyUI, some samplers or nodes might be missing. In that case, consider updating or replacing them.

MixMod

https://github.com/kantsche/ComfyUI-MixMod

extramodels

https://github.com/city96/ComfyUI_ExtraModels

GGUF loder

https://github.com/city96/ComfyUI-GGUF

T5 gguf or fp8 or fp16

https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf

https://huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/blob/main/text_encoders/t5xxl_fp8_e4m3fn.safetensors

https://huggingface.co/theunlikely/t5-v1_1-xxl-fp16/tree/main