Using checkpoint types in X/Y/Z plots appears to perform incomplete image generation.
Checkpoint A, Checkpoint B, Checkpoint C, . are listed in the Stable Diffusion UI.
The expected behaviour when that cue is executed is the generation of image A, image B, image C, .
However, in practice, when that queue is executed, an image of A, an image of A, an image of . image of A, is generated.
It is difficult to see the cause of this, but in the command line log, checkpoint A, checkpoint B, checkpoint C, . appear to be switching.
However, the checkpoints are not actually switched.
If the user manually unloaded the checkpoints from the VRAM, the image would be generated correctly.
However, the user had to repeat this action once per job.
Please consider correcting this so that the checkpoints can be switched successfully.
Hello, thank you for taking over and creating this extension.
I use X/Y/Z plots frequently and am reporting this issue. Could you please fix this one?
StableDiffusion environment where the problem is occurring.
Version: v1.3.2 Commit hash: baf6946e06249c5af9851c60171692c44ef633e0
The problem is as follows
Using checkpoint types in X/Y/Z plots appears to perform incomplete image generation.
Checkpoint A, Checkpoint B, Checkpoint C, . are listed in the Stable Diffusion UI. The expected behaviour when that cue is executed is the generation of image A, image B, image C, .
However, in practice, when that queue is executed, an image of A, an image of A, an image of . image of A, is generated.
It is difficult to see the cause of this, but in the command line log, checkpoint A, checkpoint B, checkpoint C, . appear to be switching. However, the checkpoints are not actually switched.
If the user manually unloaded the checkpoints from the VRAM, the image would be generated correctly. However, the user had to repeat this action once per job.
Please consider correcting this so that the checkpoints can be switched successfully.