receyuki / comfyui-prompt-reader-node

The ultimate solution for managing image metadata and multi-tool compatibility. ComfyUI node version of the SD Prompt Reader
MIT License
289 stars 22 forks source link

[BUG] - Image batches failing #67

Open RangerFrank opened 7 months ago

RangerFrank commented 7 months ago

Description

Image batches failing due to random seed changing from -1 to the last generated seed randomly, messing up the rest of the queue and throwing a ksampler error indicating it was fed a negative integer instead of a seed value.

Reproduction steps

Queue 6+ images on random seed, by about the 5th generation the seed will change from -1 to a fixed seed.

Image file

No response

receyuki commented 7 months ago

I can't reproduce this issue. Could you please post your workflow here and tell me how you operated it?

RangerFrank commented 7 months ago

After further testing it occurs when I change the positive prompt while there are images queued, and then queue more. The first image queued after changing the prompt will cause SD Param Gen to fix the seed to the last image in queue.

receyuki commented 7 months ago

The bad news is I still can't trigger this bug. When the error occurs, does the third button under Param Gen display the correct number? Have you tried deleting the node, refreshing the page, and re-adding this node?

RangerFrank commented 7 months ago

I have tried readding and rebuilding the node. I'm starting to think it might be a parsing issue with Efficiency Nodes KSampler

RangerFrank commented 7 months ago

The problem occurs when queued image settings change. If you keep the seed fixed you can queue images with different conditioning, loras, etc without issue. If you randomise the seed and queue a number of images, changing any of the settings and queuing 2 or more will cause the seed to become fixed when the first different generation hits the ksampler and following queued generations will fail.

I've found a workaround for this by using the Global Seed node from Inspire Pack, this allows for seed generation before queueing instead of after, fixing the issue. It might be good to have some of that functionality built directly into the parameter generator node.