WAN2.2_I2I_upscale

V 1.0-1.1

A WAN upscaling WF that with a standard 832x1216 image takes about 150-180 secs on my 12GB 3060 with 32GB RAM. I tried to keep it as simple as possible. Lower denoise for end-results closer to the original, increase for more creative ones. I use it to improve the overall quality of Chroma and Qwen-img low step/reduced resolution images (much faster generation).

V 1.2

KJ Sage attention patch node stopped working at comfy version 0.3.59. If you select v. 0.3.58 it still works. Alternatively you can either bypass it (ctrl+B) and load sage attention at startup but that still causes the black picture bug with Qwen-image. On a different note this i2i wf does a better job at preserving the original's features at low denoise (0.15-0.2) and improves the overall image quality. It uses the quantized version of wan2.2-t2v-rapid-aio-v10-nsfw.safetensors that can be found here: https://huggingface.co/befox/WAN2.2-14B-Rapid-AllInOne-GGUF/tree/main/v10

Depending on how much VRAM you have download the appropriate version.