Skip to content

[Bug]: Interpolation resume error #5

@optisynapsis

Description

@optisynapsis

What happened and why was it ungood?

I have my own modified version of render_modes.py which I adapted for personal use,

but it seems in the current version, that resuming an interpolation render doesn't work properly in sd-forge-deforum.

The bug seems to be in line 163 :
opts.data["CLIP_stop_at_last_layers"] = scheduled_clipskip if scheduled_clipskip is not None else opts.data["CLIP_stop_at_last_layers"]

Commenting this out means the error isn't thrown, but the resume function still doesn't appear to work correctly and starts from the first frame, if the resume is attempted again.

(I'm curious if this will have any effect on the output if Clip Skip schedule is not enabled).

My own modified from sd-forge-deforum render_modes.py works with resume capabilities.

It can resume a render, accounting for missing frames, but it also does some unnecessary things, like writing out an empty text file with the exact cfg value for each image in the sequence (these seem to be very likely to have a high number of significant digits, with an unpredictably high variance between very small changes).

However, the resume loop is a little more versatile, and seems to be working, so it might be worth implementing officially.

Obviously it'd be more reasonable to find the cause of the aforementioned bug.

Not too familiar with git, so unlikely to be able to submit a fix myself, but that bug should probably be looked into first before my more 'elegant' resume solution could be considered.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions