Add slice_cond and per-model context window cond resizing (#12645)
* Add slice_cond and per-model context window cond resizing * Fix cond_value.size() call in context window cond resizing * Expose additional advanced inputs for ContextWindowsManualNode Necessary for WanAnimate context windows workflow, which needs cond_retain_index_list = 0 to work properly with its reference input. ---------
This commit is contained in:
@@ -285,6 +285,12 @@ class BaseModel(torch.nn.Module):
|
||||
return data
|
||||
return None
|
||||
|
||||
def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
|
||||
"""Override in subclasses to handle model-specific cond slicing for context windows.
|
||||
Return a sliced cond object, or None to fall through to default handling.
|
||||
Use comfy.context_windows.slice_cond() for common cases."""
|
||||
return None
|
||||
|
||||
def extra_conds(self, **kwargs):
|
||||
out = {}
|
||||
concat_cond = self.concat_cond(**kwargs)
|
||||
@@ -1375,6 +1381,12 @@ class WAN21_Vace(WAN21):
|
||||
out['vace_strength'] = comfy.conds.CONDConstant(vace_strength)
|
||||
return out
|
||||
|
||||
def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
|
||||
if cond_key == "vace_context":
|
||||
import comfy.context_windows
|
||||
return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=3, retain_index_list=retain_index_list)
|
||||
return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
|
||||
|
||||
class WAN21_Camera(WAN21):
|
||||
def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
|
||||
super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.CameraWanModel)
|
||||
@@ -1427,6 +1439,12 @@ class WAN21_HuMo(WAN21):
|
||||
|
||||
return out
|
||||
|
||||
def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
|
||||
if cond_key == "audio_embed":
|
||||
import comfy.context_windows
|
||||
return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=1)
|
||||
return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
|
||||
|
||||
class WAN22_Animate(WAN21):
|
||||
def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
|
||||
super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model_animate.AnimateWanModel)
|
||||
@@ -1444,6 +1462,14 @@ class WAN22_Animate(WAN21):
|
||||
out['pose_latents'] = comfy.conds.CONDRegular(self.process_latent_in(pose_latents))
|
||||
return out
|
||||
|
||||
def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
|
||||
import comfy.context_windows
|
||||
if cond_key == "face_pixel_values":
|
||||
return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=2, temporal_scale=4, temporal_offset=1)
|
||||
if cond_key == "pose_latents":
|
||||
return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=2, temporal_offset=1)
|
||||
return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
|
||||
|
||||
class WAN22_S2V(WAN21):
|
||||
def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
|
||||
super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.WanModel_S2V)
|
||||
@@ -1480,6 +1506,12 @@ class WAN22_S2V(WAN21):
|
||||
out['reference_motion'] = reference_motion.shape
|
||||
return out
|
||||
|
||||
def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
|
||||
if cond_key == "audio_embed":
|
||||
import comfy.context_windows
|
||||
return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=1)
|
||||
return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
|
||||
|
||||
class WAN22(WAN21):
|
||||
def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
|
||||
super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.WanModel)
|
||||
|
||||
Reference in New Issue
Block a user