Fix batch handling, attempt to deal with non-4 dimensional latents - #6
Merged
Merged
Conversation
Owner
|
Oh wow! Thank you!! |
Owner
|
I'd be curious on how to implement and ancestral version BTW! I only gave this a weak try and got more noise at the end 😅 |
Contributor
Author
I've actually had a working version (though based on an older version of Distance) in my Overly Complicated Sampling node pack for a while now: https://github.com/blepping/comfyui_overly_complicated_sampling/blob/heun_experiments/py/step_samplers/extraltodeus.py I created a pull that adds ancestral support over in #7. |
Owner
|
Well it's impressive! Thank you very much! :D |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
As noted in #5, DistanceSampler currently fails on batches. This pull implements the simplest/dumbest solution: we just cram the batch (and any extra dimensions) into the channel dimension. I don't know if this is a correct solution, but it does seem to work and produce reasonable results. These changes also fix DistanceSampler for non-4 dimensional (3+) dimensions. For example, video models like Wan use 5 dimensional latents. Theoretically these changes would also work with 3 dimensional latents but I don't know of any 1D VAE models to test with at the moment. I did test with Wan and it works, even with batches of video latents.
Should close #5
Not sure if you're interested but I can also submit a pull that allows setting the various DistanceSampler options in the node rather than having to configure them by preset:
Would also be pretty easy to add ancestral sampling support. Let me know if you'd accept a pull adding that kind of stuff!