-
Notifications
You must be signed in to change notification settings - Fork 6.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Preview sampled images with TAESD #713
Preview sampled images with TAESD #713
Conversation
This is great. Definitely a great core feature. Though, why is the sampler preview version so much richer in color? |
This feature will be a powerful impression for users looking at ComfyUI. |
Yeah it's clamped beforehand, I think some of the code I adapted from A1111 is suspect... Some of it unscales latents to a range within |
After testing this a bit I don't think having a previewer node connection is a good idea. Since previews are not something that affect the output image they should not be a part of the workflow itself. I think to start previews should just be a command line option: |
I agree with the first part, but not this. You may want a preview for something like a taxing HR image, but not something like a small 512x512 hand fix, or face fix. Simple |
Is there a use case for some nodes having previews and others not/having less expensive preview methods? Or being able to individually change settings for each node's preview? Also this still doesn't help with the case of nodes that want custom preview methods besides |
I literally just gave one. You may not want to slow down a process via callback previews if it's not necessary, especially if you're not caring about the result as something you may want to cancel. And if isn't about whether or not the result is worth continuing, it's not really a useful function (in general) at all besides "ooh,look, the diffusion process" |
I followed the instructions but still don't understand. Where's the TAESD Decode node from your screenshot? |
Lets KSampler show previews with taesd
Download the
.pth
models from https://github.com/madebyollin/taesd and stick them inmodels/taesd
firstPreviews are encoded to JPEG bytes first then sent over websockets, they're resized to 512 pixels first so the network load shouldn't be too much (~50kb per frame for 2048x2048 resolution)
Also adds taesd-related nodes, they're functionally equivalent to VAE loader/encoder/decoder nodes
output2.webm