light-and-ray/sd-webui-old-sd-firstpasser

Firstpass generation with old SD model, Loras, embedding, etc

sd-webui-old-sd-firstpasser's Issues

It would be nice if the prompt could be separated separately.

since there is a characteristic difference between the prompt when running the sd 1.5 model and the prompt when running on the SDXL, it would be nice to add a separate prompt window to the script so that the two prompts can be written separately.

Separating the firstpass size from the SDLX one

The script use an upscaler, however, there's currently no genuine upscale process since the sizes of the first pass and the finale results are the same.

Issue with SD3

The Auto setting didn't work (Exception: Can't guess 'Firstpass network type', please set it up manually), so I tried SDXL, and I get this error:

Error in UI:

RuntimeError: Error(s) in loading state_dict for LatentDiffusion: size mismatch for first_stage_model.encoder.conv_out.weight: copying a param with shape torch.Size([32, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 512, 3, 3]). size mismatch for first_stage_model.encoder.conv_out.bias: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([8]). size mismatch for first_stage_model.decoder.conv_in.weight: copying a param with shape torch.Size([512, 16, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 4, 3, 3]).

Error in CLI:

File "/home/hunter/Projects/ai/stable-diffusion-webui-debug/modules/sd_disable_initialization.py", line 223, in <lambda>
        module_load_state_dict = self.replace(torch.nn.Module, 'load_state_dict', lambda *args, **kwargs: load_state_dict(module_load_state_dict, *args, **kwargs))
                                                                                                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/home/hunter/Projects/ai/stable-diffusion-webui-debug/modules/sd_disable_initialization.py", line 221, in load_state_dict
        original(module, state_dict, strict=strict)
      File "/home/hunter/Projects/ai/stable-diffusion-webui-debug/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 2191, in load_state_dict
        raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
    RuntimeError: Error(s) in loading state_dict for LatentDiffusion:
    	size mismatch for first_stage_model.encoder.conv_out.weight: copying a param with shape torch.Size([32, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 512, 3, 3]).
    	size mismatch for first_stage_model.encoder.conv_out.bias: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([8]).
    	size mismatch for first_stage_model.decoder.conv_in.weight: copying a param with shape torch.Size([512, 16, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 4, 3, 3]).

Firstpass model is stableDiffusion3SD3_sd3Medium.safetensors

need to fix

#1

1. vae in override, it will create image with strange colors
2. seed -1 (random), I belive what's happening is that the seed used across different passes of image gen is random and not recorded in infotext, as such the information is lost and you won't be able to recreate the image
3. fix txt2img progressbar with hires fix