Lachlan Nicholson

I solved a similar error by using a different torch cuda SDP backend. Specifically the math/pytorch c++ implementation worked fine, but flash and memory-efficient attention failed. More info here: https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html

Definitely possible, I might try and get at this shortly. The noise parameters for the boundingbox factors are crucial for system performance. In the paper I describe a simple approach...

At a quick glance, that dual conic doesn't seem to be well defined. The mathematics for projecting dual quadrics should be well defined even at edge cases, but perhaps there...

The TransparentVAEDecoder is compatible with Diffusers, though the LoRA models may take some effort in translating keys.

@arpitsahni04 Specifically for the core LoRA model (not the others which change layer sizes): Diffusers have some support for converting SGM/Automatic/Kohya format loras to diffusers format. This lora seemed to...

@pin-lpt-heatherwick I can confirm it does work. Those shapes you shared are okay - the first is an up weight and the second is a down weight, so there shapes...