Open Kallamamran opened 4 days ago
Try reverting ComfyUI to an earlier version and replicate the results. goto ComfyUI dir in cmd. And execute this
git checkout v0.2.1
It's necessary to determine whether the issue truly lies with ComfyUI or with external environment like PyTorch.
Can I go back to the new version then? I really don't want to mess up mu installation
This also happened to me last year when I updated A1111. Most likely pytorch or some other library got updated and the calculations are slightly different now. So, the images are different as well.
Can I go back to the new version then? I really don't want to mess up mu installation
You can go back using this command
git checkout master
If it is a PyTorch issue, you can resolve it by downgrading your PyTorch version.
Expected Behavior
I expect it to generate the same image if I drag in the image including the workflow and click generate. No changes in hardware. Same models used. The only thing that has changed since August is ComfyUI updates.
Actual Behavior
Images generated August 15 and now is different. Not by a lot, but by enough to get the feeling something is wrong. The images generated now compared to back in August is missing something. It's like they have lost their soul or something. Images to the left are generated back in August. Images to the right are generated now. Judge for yourself. It's like ComfyUI makes the model forget how to turn tires, how to do motionblur, how to colourize and they also get that pixelated fram-line that can turn up when an image is not divisible by a certain number... I don't know, but something is seriously missing. Something is wrong!!!
Steps to Reproduce
Generate an image in ComfyUI installed and not updated since August 15, 2024 Compare it to the same image generated in an updated ComfyUI = 😥
Debug Logs
Other
No response