Original model:https://civarchive.com/models/50088/angra-realflex?modelVersionId=246988
merged SD 1.5 LCM weights in at 1 strength for speed using LCM sampler, 1 cfg, 5-15 steps
Description
FAQ
Comments (5)
Model it does not work
I used it for the examples, and it's just the base model with LCM weights merged in - identical to running the source model and LCM lora at weight 1. It's not my model - if the original plus LCM lora also doesn't work then something else is wrong.
@CautionMergeAhead
Yeah I had issues too. Using sd.next. didn't try solving it but this is what I got if anyone else interested :)
ValueError: Cannot load because text_model.embeddings.position_embedding.weight expected shape tensor(..., device='meta', size=(77, 1280)), but got torch.Size([77, 768]). If you want to instead overwrite randomly initialized weights, please make sure to pass both low_cpu_mem_usage=False and ignore_mismatched_sizes=True. For more information, see also: https://github.com/huggingface/diffusers/issues/1619#issuecomment-1345604389 as an example.
It definitely does work, because I get normal images with LCM sampler and Diffusers pipeline.
But value of model is low - it is 3.6Gb in size but output is almost identical to any other "photorealistic" merge which are 2 Gb size. It dos not have own style, it does not add any details, it does not produce higher image quality. (And it is not faster than others, generation time exactly the same as others photorealistic).
@littlefluffyball Thanks for the feedback. I literally just merged LCM weights with the model and posted. I did nothing to adjust anything, and the base model wasn't mine. I believe I was clear in the notes about it, so I won't be doing anything with this.
If anyone has issues with this, then just download the AngrA model (they probably have a newer version now) and the LCM weight lora and do that - this is no better.