Cushy's Stupendous Image Upscaler

From 0.3 to 256 MPixels! Or from at least 0.3 to 32 MPixel in ONE MINUTE!

Congratulations! The fact you reading this, means you are probably not stupid. How clever of you. So have a little read.

One day, I felt more bad then normal, I felt more bored then I normally do by life and the so called Civilization. So I sat here and stared at my stupid screen with a stupid face expression and started kind of drifting away into a awake dreaming state.

I played a lot with Flux 9 back in May when it was new and hot. And I had made some tests with upscaling, but it all sucked. So, while I was sitting on by uncomfortable chair, and was a little drooling, I had an Idea:

Why not try again. Yes! Why not create the Cushy's stupendous upscaler, that upscales like no other upscaler before!? So I tried and tried, an mixed and went back a step and went forward one, added another node, deleted one, played on that dark canvas, drank at least 1 liter beer, and there it was, a very overdone upscaler. I was so proud!

But what does it matter if you can't elevate your little ego over you oppo..., sorry, - friends to make them look small and incompetent...

One week later or less someone with actual intelligence released the flow upscaler, and I was put me to my place. "But not with me!" thought my pesky little brain, - let's combine our works and make it absolutely ridiculous!

And so I did. And here we are, I hope you will have fun!

Ok Cushy, but what does this thing exactly do?

It will enhance any given Image from 0.3 to 256 MPixels! Or more realistic, from at least 0.3 to 32 Mpixel in ONE MINUTE! And it will look very good, - at least some times!

But why?

You may know why. Maybe for a print, or a remaster, or a reconstruction or what is called a re-imagination.

Ok, But what does it need?

• ComfyUI

• The attached workflow
  

• A Flux 9 klein variant, it works fine with NVFP4 ( e.g. ungated here: https://huggingface.co/ApacheOne/FluxKlein9b-nvfp4_mixed/tree/main ) (to models\diffusion_models)

• https://huggingface.co/datasets/zhenzi/comfyui-models/blob/main/models/text_encoders/qwen_3_8b_fp8mixed.safetensors (to models\text_encoders)

• https://huggingface.co/Comfy-Org/flux2-dev/blob/main/split_files/vae/flux2-vae.safetensors (to models\vae)

After loading my workflow, you need to download the missing nodes with the manager, click on "Install Missing Custom Nodes" to get them. It may work for models too (Model Manager). Or do it manually.

If you want or need to manual downloading models and nodes:

You need this node:

• https://github.com/tensorforger/comfyui-flow-upscaler

• Place it in ComfyUI/custom_nodes

You need this weights:

• https://huggingface.co/TensorForger/FlowUpscaler

• Place it in ComfyUI/models/latent_upscale_models

Download this light-weight VAE

• https://huggingface.co/madebyollin/taef2/blob/main/taef2.safetensors

• Place it in ComfyUI/models/vae_approx

Download this upscaler model

• https://huggingface.co/Acly/Omni-SR/blob/main/OmniSR_X3_DIV2K.safetensors

• Place it in ComfyUI/models/upscale_models

Download this node pack:

• https://github.com/cubiq/ComfyUI_essentials

• Place it in ComfyUI/custom_nodes

Download this node pack:

• https://github.com/capitan01R/ComfyUI-Flux2Klein-Enhancer

• Place it in ComfyUI/custom_nodes

Download this node pack:

• https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUI

• Place it in ComfyUI/custom_nodes

You know what, - USE THE DAMN MANAGER! - Jokes aside, you should be ready to go. Enhance!