Control Net Upscaler for Z-image Turbo?

#9
by BetinhoFormado - opened

I was wondering if we could have something like this control net for the z-image? It would be great to get faster results than we do with Flux. Also, this can be used not only for upscaling for a better solution than img2img by keeping very close to the source and adding details

https://huggingface.co/jasperai/Flux.1-dev-Controlnet-Upscaler

Alibaba-PAI org

I'd be happy to work on this. Do you mean inputting low-quality images and generating high-resolution outputs? What target resolution are you aiming for?

Yes pretty much helps upscale like this link examples/showcase.jpg on the hugging space link I posted pretty much that get low quality images like compressed jpgs, grain, low resolution and creates a better image on the full sampling size.

Hi you can use Seedvr2 for upscale , works very well with z-image ;)

Jackburton, I don't think you understood the usage here. It is not about getting big. I have already been using SeedVR2.5 for a while. This is for fixing very low-resolution images, but it's also excellent for creative upscaling, producing results similar to Magnific AI Research's divide-and-conquer workflow. You will understand. This process allows you to upscale to more than 10k images with Creative Upscale. No Sweat SeedVR2.5 has a memory limit for large images. You can use 5090, max 8k, and it is not a creative upscaler; it doesn't work in some cases. It's just that clarity upscaler and magnific AI use the same method, but FRE,E with the speed and quality of z-image, will allow anyone to get 10k images easily with an 8GB GPU with GGUF. And you welcome those with low VRAM and didn't know this exists

Sign up or log in to comment