diff options
Diffstat (limited to 'modules/sd_samplers_kdiffusion.py')
| -rw-r--r-- | modules/sd_samplers_kdiffusion.py | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/modules/sd_samplers_kdiffusion.py b/modules/sd_samplers_kdiffusion.py index 528f513f..93f0e55a 100644 --- a/modules/sd_samplers_kdiffusion.py +++ b/modules/sd_samplers_kdiffusion.py @@ -101,11 +101,13 @@ class CFGDenoiser(torch.nn.Module): sigma_in = torch.cat([torch.stack([sigma[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [sigma] + [sigma])
image_cond_in = torch.cat([torch.stack([image_cond[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [image_cond] + [torch.zeros_like(self.init_latent)])
- denoiser_params = CFGDenoiserParams(x_in, image_cond_in, sigma_in, state.sampling_step, state.sampling_steps)
+ denoiser_params = CFGDenoiserParams(x_in, image_cond_in, sigma_in, state.sampling_step, state.sampling_steps, tensor, uncond)
cfg_denoiser_callback(denoiser_params)
x_in = denoiser_params.x
image_cond_in = denoiser_params.image_cond
sigma_in = denoiser_params.sigma
+ tensor = denoiser_params.text_cond
+ uncond = denoiser_params.text_uncond
if tensor.shape[1] == uncond.shape[1]:
if not is_edit_model:
|