nodes/kijai-ComfyUI-HunyuanVideoWrapper
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fixed model fixes

Browse Source
This commit is contained in:
kijai
2025-03-07 22:54:18 +02:00
parent d4d93377e5
commit 0935213be2
1 changed files with 5 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hyvideo/diffusion/pipelines/pipeline_hunyuan_video.py
+5 -7
View File
@@ -684,6 +684,9 @@ class HunyuanVideoPipeline(DiffusionPipeline):
if self.interrupt: if self.interrupt:
continue continue
if image_cond_latents is not None and i2v_condition_type == "token_replace":
latents = torch.concat([original_image_latents, latents[:, :, 1:, :, :]], dim=2)
latent_model_input = latents latent_model_input = latents
input_prompt_embeds = prompt_embeds input_prompt_embeds = prompt_embeds
input_prompt_mask = prompt_mask input_prompt_mask = prompt_mask
@@ -730,8 +733,8 @@ class HunyuanVideoPipeline(DiffusionPipeline):
t_expand = t.repeat(latent_model_input.shape[0]) t_expand = t.repeat(latent_model_input.shape[0])
#if leapfusion_img2vid: if leapfusion_img2vid:
# latent_model_input[:, :, [0,], :, :] = original_latents[:, :, [0,], :, :].to(latent_model_input) latent_model_input[:, :, [0,], :, :] = original_latents[:, :, [0,], :, :].to(latent_model_input)
if image_cond_latents is not None and not use_context_schedule: if image_cond_latents is not None and not use_context_schedule:
if i2v_condition_type == "latent_concat": if i2v_condition_type == "latent_concat":
@@ -739,11 +742,6 @@ class HunyuanVideoPipeline(DiffusionPipeline):
i2v_mask = torch.cat([i2v_mask] * 2) if cfg_enabled else i2v_mask i2v_mask = torch.cat([i2v_mask] * 2) if cfg_enabled else i2v_mask
latent_image_input = torch.cat([latent_image_input, i2v_mask], dim=1) latent_image_input = torch.cat([latent_image_input, i2v_mask], dim=1)
latent_model_input = torch.cat([latent_model_input, latent_image_input], dim=1) latent_model_input = torch.cat([latent_model_input, latent_image_input], dim=1)
elif i2v_condition_type == "token_replace" or leapfusion_img2vid:
latent_image_input = (torch.cat([original_image_latents] * 2) if cfg_enabled else original_image_latents)
latent_model_input = torch.cat([latent_image_input, latent_model_input[:, :, 1:, :, :]], dim=2)
else:
latent_model_input = torch.cat([latent_model_input, latent_image_input], dim=1)
if self.transformer.guidance_embed: if self.transformer.guidance_embed:
if cfg_enabled: if cfg_enabled: