Img2Img and extras upscaler like a1111 in Comfyui版本v1.0 (ID: 296327)

I have just starting to use Comfyui recently and wasn't satisfied with the up-scaling method I learn from the goat Mr.Olivio S. because :

• the up-scaling models set by default tend to add much more details than the low resolution image and change a bit the style of the model

• They need at least 0.40-0.50 denoising

• I feel like they may be very useful for realistic art or semi-realistic/2.5D but not very interesting for simple detailed anime images.

Since I produce only anime like outputs, I try with the few knowledge I get from Olivio's Workshop to find again the a1111 up-scaling method.

I could achieve this simple workflow in the attachment.

I found that is very similar to how we upscale an image in a1111, see comparison in posts attached to the article.

The closer from 0.00 you are in the "Latent blend" "blend_factor", the closer you are from a traditional a1111 upscale. I recommend between 0.00 and 0.20. Too high, it changes the style.

It is set up here to upscale images doing 512x768 pixels to 768x1152 pixels (x1.50) to 1536x2304 pixels (extras x2). To change those value, you need to change values in latent empty images and width in the CR upscale.

As I am just a beginner with comfy, if you have better workflow for this purpose, I'd like to know, you can share it in the comments.

Still learning comfy, looking forward to build new original workflows.

Comparison of a1111 up-scaling process and my workflow (same settings and denoising) in posts section.

描述:

First version, could be upgrade

训练词语: