search

ttulttul / ComfyUI-Iterative-Mixer

Nodes that implement iterative mixing of samples to help with upscaling quality
GNU General Public License v3.0
115 stars 9 forks source link

Hello, could you provide any other more intuitive examples? #18

Open shawpook opened 10 months ago

shawpook commented 10 months ago

I tried your workflow and I can see its function in it, but when I tried it in my own workflow, it only generated a lot of noise. The difference between my workflow and yours is that I am not used to using LCM Lora and sampler. I used DPM and also tried Eular, and the results are far from the example. Is this node only suitable for LCM to use? Sorry, I just tried my best to optimize the quality of the images. I am not very familiar with the technical aspects. I can only understand a little of your text. If my problem bothers you, I hope you can forgive me

ttulttul commented 10 months ago

I will try to provide some more helpful workflows demonstrating proven combinations of settings. I may also create an "ease" node that offers known-good options.

The only sampler choices that I know work well with the IterativeMixingSampler node are the euler and euler_perlin. I do have basic implementations for LCM and dpm2m-pp, but I have not gotten good results from them yet.

Stay tuned to the repo and watch for those better examples soon.

shawpook commented 10 months ago

I will try to provide some more helpful workflows demonstrating proven combinations of settings. I may also create an "ease" node that offers known-good options.

The only sampler choices that I know work well with the IterativeMixingSampler node are the euler and euler_perlin. I do have basic implementations for LCM and dpm2m-pp, but I have not gotten good results from them yet.

Stay tuned to the repo and watch for those better examples soon.

I'm looking forward to it, thank you very much

ttulttul commented 9 months ago

Check out the new inpainting example. I have been working hard on inpainting with the iterative mixer lately and have been getting really exceptional results. More coming soon.

shiloh92 commented 9 months ago

Updating my comment: It took 38 mins on 4GB VRAM to complete the iterative sampler demonstration, just an FYI to anyone with modest machines. I used euler_perlin. The results were unclear to me whether an improvement or not. Should I try the LCM or dpm setting instead?

ttulttul commented 8 months ago

Sorry for the delay in writing back. Yes, that workflow is massive and I don't know what to say other than perhaps this is 4090 territory.

I would avoid the Perlin noise feature; it doesn't seem to add much. The rewind setting that was added recently actually does accomplish adding detail and I'm focusing all my efforts on exploring that now to see how far we can push it.