2025-10-03 07:47:56 - main - INFO - loading adapter detail_enhancer
No LoRA keys associated to CLIPTextModel found with the prefix='text_encoder'. This is safe to ignore if LoRA state dict didn't originally have any CLIPTextModel related params. You can also try specifying prefix=None to resolve the warning. Otherwise, open an issue if you think it's unexpected: https://github.com/huggingface/diffusers/issues/new
No LoRA keys associated to CLIPTextModelWithProjection found with the prefix='text_encoder_2'. This is safe to ignore if LoRA state dict didn't originally have any CLIPTextModelWithProjection related params. You can also try specifying prefix=None to resolve the warning. Otherwise, open an issue if you think it's unexpected: https://github.com/huggingface/diffusers/issues/new

This warning confirms what we've been discussing: detail_enhancer (and likely nijistyle and chibi_rr) only have UNet weights, not text encoder weights.

The warnings are saying the LoRA files don't contain any parameters for the text encoders - they're UNet-only adapters.
Why This Causes Issues After Compilation

When you call pipeline.set_adapters(['detail_enhancer']):

Without compilation:

Pipeline checks each component Applies adapter to UNet ✓ Skips text encoders (no weights found) ✓ Works fine 

With compilation:

The compiled graph expects a consistent structure If you compile after loading adapters that ARE in text encoders, the graph expects text encoder adapters When you try to use detail_enhancer which has no text encoder weights, the compiled code path fails