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

implement tangential guidance

Browse Source
This commit is contained in:
kabachuha
2025-07-28 21:08:10 +03:00
parent 998a69cc0a
commit fd285a7e7d
2 changed files with 21 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nodes.py
+11 -4
View File
@@ -12,7 +12,7 @@ from .fp8_optimization import convert_linear_with_lora_and_scale, remove_lora_fr
from .wanvideo.schedulers import get_scheduler, get_sampling_sigmas, retrieve_timesteps, scheduler_list
from .gguf.gguf import set_lora_params
from .multitalk.multitalk import timestep_transform, add_noise
from .utils import log, print_memory, apply_lora, clip_encode_image_tiled, fourier_filter, is_image_black, add_noise_to_reference_video, optimized_scale, setup_radial_attention, compile_model, dict_to_device
from .utils import log, print_memory, apply_lora, clip_encode_image_tiled, fourier_filter, is_image_black, add_noise_to_reference_video, optimized_scale, setup_radial_attention, compile_model, dict_to_device, tangential_projection
from .cache_methods.cache_methods import cache_report
from .enhance_a_video.globals import set_enhance_weight, set_num_frames
from .taehv import TAEHV
@@ -1221,6 +1221,7 @@ class WanVideoExperimentalArgs:
"fresca_scale_low": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 10.0, "step": 0.01}),
"fresca_scale_high": ("FLOAT", {"default": 1.25, "min": 0.0, "max": 10.0, "step": 0.01}),
"fresca_freq_cutoff": ("INT", {"default": 20, "min": 0, "max": 10000, "step": 1}),
"use_tcfg": ("BOOLEAN", {"default": False, "tooltip": "https://arxiv.org/abs/2503.18137 TCFG: Tangential Damping Classifier-free Guidance. CFG artifacts reduction."}),
},
}
@@ -1881,7 +1882,7 @@ class WanVideoSampler:
timesteps[-drift_steps:] = drift_timesteps[-drift_steps:]
# Experimental args
use_cfg_zero_star = use_fresca = False
use_cfg_zero_star = use_tangential = use_fresca = False
if experimental_args is not None:
video_attention_split_steps = experimental_args.get("video_attention_split_steps", [])
if video_attention_split_steps:
@@ -1891,6 +1892,7 @@ class WanVideoSampler:
use_zero_init = experimental_args.get("use_zero_init", True)
use_cfg_zero_star = experimental_args.get("cfg_zero_star", False)
use_tangential = experimental_args.get("use_tcfg", False)
zero_star_steps = experimental_args.get("zero_star_steps", 0)
use_fresca = experimental_args.get("use_fresca", False)
@@ -2174,6 +2176,11 @@ class WanVideoSampler:
).view(batch_size, 1, 1, 1)
else:
alpha = 1.0
noise_pred_uncond_scaled = noise_pred_uncond * alpha
if use_tangential:
noise_pred_uncond_scaled = tangential_projection(noise_pred_cond, noise_pred_uncond_scaled)
#https://github.com/WikiChao/FreSca
if use_fresca:
@@ -2183,9 +2190,9 @@ class WanVideoSampler:
scale_high=fresca_scale_high,
freq_cutoff=fresca_freq_cutoff,
)
noise_pred = noise_pred_uncond * alpha + cfg_scale * filtered_cond * alpha
noise_pred = noise_pred_uncond_scaled + cfg_scale * filtered_cond * alpha
else:
noise_pred = noise_pred_uncond * alpha + cfg_scale * (noise_pred_cond - noise_pred_uncond * alpha)
noise_pred = noise_pred_uncond_scaled + cfg_scale * (noise_pred_cond - noise_pred_uncond_scaled)
return noise_pred, [cache_state_cond, cache_state_uncond]
utils.py
+10
View File
@@ -501,3 +501,13 @@ def compile_model(transformer, compile_args=None):
transformer = torch.compile(transformer, fullgraph=compile_args["fullgraph"], dynamic=compile_args["dynamic"], backend=compile_args["backend"], mode=compile_args["mode"])
return transformer
def tangential_projection(pred_cond: torch.Tensor, pred_uncond: torch.Tensor) -> torch.Tensor:
cond_dtype = pred_cond.dtype
preds = torch.stack([pred_cond, pred_uncond], dim=1).float()
orig_shape = preds.shape[2:]
preds_flat = preds.flatten(2)
U, S, Vh = torch.linalg.svd(preds_flat, full_matrices=False)
Vh_modified = Vh.clone()
Vh_modified[:, 1] = 0
recon = U @ torch.diag_embed(S) @ Vh_modified
return recon[:, 1].view(pred_uncond.shape).to(cond_dtype)