Image upscale pipeline

[saddam213]

Add a new pipeline for image up-scaling

• [x] Upscale Diffuser
• [x] UI View for up-scaling
• [ ] Integrate with existing pipelines
• [ ] Split images into tiles and batch process
• [ ] Model Loader to support models with vae folder and no vae_encoder/vae_decoder folders

Will be using the following model for development https://huggingface.co/ssube/stable-diffusion-x4-upscaler-onnx

[saddam213]

Upon further investigation into implementation it seems the max sample size for upscale is 128

128 -> 512

For images over 128 we are going to have to split into tiles, upscale and reassemble

512 -> [128, 128, 128, 128] -> [512, 512, 512, 512] -> 2048

[Amin456789]

there is an upscaler in onnx named SwinIR, which included some models in it like RealSR [best upscaler i ever used] it is a very small onnx model that can be found here: https://huggingface.co/rocca/swin-ir-onnx/tree/main it will be great if it be included as another model for ur upscaler

[saddam213]

there is an upscaler in onnx named SwinIR, which included some models in it like RealSR [best upscaler i ever used] it is a very small onnx model that can be found here: https://huggingface.co/rocca/swin-ir-onnx/tree/main it will be great if it be included as another model for ur upscaler

Thanks for this link dude.

I have been trying for weeks to get a diffusion upscaler to work, but never got a good result I tried this one and it worked first time, plus you don't have to tile the images, I can just give it a 512 image and it spits out a 2048

However these type of upscalers do not work like diffusion ones, e.g no steps, prompt needed, so I need to think of a good way to add this to the codebase

I will add the x2 and x4 versions along with smaller ones like Face Upscalers etc

[Amin456789]

u r welcome, thanks!

[saddam213]

Closing PR, new implementation needed