Degradation in Image Quality When Using Multiple LoRAs with Flux in Forge

Nobody has claimed this yet.

I'm encountering a problem when using Flux in Forge along with dynamic prompts and multiple LoRAs. My workflow involves using wildcards to vary the output of a batch by cycling through different LoRAs. However, I’ve noticed that the quality of the images degrades as the batch progresses.

Specifically, the first few images in the batch come out great, but as the process continues, the images become increasingly pixelated and grainy. It seems like the LoRAs from the earlier images aren’t being fully unloaded, which might be causing this degradation. However, that’s just my non-expert interpretation—there could be an entirely different cause.

Could you help diagnose what might be happening here? Any insights or advice on how to resolve this would be greatly appreciated.

System info: rtx3090. On the latest commit

Start by reproducing the reported workflow in Forge: Flux with dynamic prompts, a batch cycling through multiple LoRAs, and an RTX 3090. Compare early and later images while checking whether earlier LoRAs remain loaded; done means identifying the cause and documenting or validating a resolution.

Written by the indexing model from the issue text.