ByteDance

/

SDXL-Lightning

like 1.92k

Follow

ByteDance 708

Same SDXL pipeline should work, but I haven't looked into the setting of the SDXL img2img pipeline to make sure that everything is coded up as expected.

For 1/2/4-step model, the best img2img strength is 25%, 50%, or 75%.
For 8-step model, the best img2img noise strength 12.5%, 25%, 37.5%, 50%, 62.5%, 75%, 87.5%
These settings are what the model was trained for.

1

I got this error for img2img:
Traceback (most recent call last):
File "D:\sd\test_sdxl_lighting_img.py", line 35, in
sd_model(
File "C:\Users\mi\miniconda3\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\pipelines\stable_diffusion_xl\pipeline_stable_diffusion_xl_img2img.py", line 1477, in call
image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\utils\accelerate_utils.py", line 46, in wrapper
return method(self, *args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\models\autoencoders\autoencoder_kl.py", line 302, in decode
decoded = self._decode(z).sample
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\models\autoencoders\autoencoder_kl.py", line 273, in _decode
dec = self.decoder(z)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\models\autoencoders\vae.py", line 333, in forward
sample = self.mid_block(sample, latent_embeds)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\models\unets\unet_2d_blocks.py", line 660, in forward
hidden_states = self.resnets[0](hidden_states, temb)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\diffusers\models\resnet.py", line 338, in forward
hidden_states = self.nonlinearity(hidden_states)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\modules\activation.py", line 393, in forward
return F.silu(input, inplace=self.inplace)
File "C:\Users\mi\miniconda3\lib\site-packages\torch\nn\functional.py", line 2072, in silu
return torch._C._nn.silu(input)
TypeError: silu(): argument 'input' (position 1) must be Tensor, not NoneType

Wow, for my vae, strength should be equal to or larger than 0.5.