Repository navigation
WanVideoToVideoPipeline: "No cache context is set" when caching is enabled #15006
Description
Describe the bug
Enabling FirstBlockCache, TaylorSeer, or MagCache on WanVideoToVideoPipeline fails on the first denoising step with ValueError: No cache context is set. The conditional and unconditional transformer calls in the pipeline don't run inside a cond / uncond cache context.
This is the Wan video-to-video case of #12760.
Reproduction
import torch
from diffusers import WanVideoToVideoPipeline, FirstBlockCacheConfig
from diffusers.utils import load_video
pipe = WanVideoToVideoPipeline.from_pretrained(
"Wan-AI/Wan2.1-T2V-1.3B-Diffusers", torch_dtype=torch.bfloat16
)
pipe.to("cuda")
pipe.transformer.enable_cache(FirstBlockCacheConfig(threshold=0.2))
video = load_video("<any input video>")
out = pipe(video=video, prompt="a cat walking", strength=0.5, guidance_scale=5.0,
num_inference_steps=16, height=480, width=832).frames[0]Logs
ValueError: No cache context is setSystem Info
diffusers main (@ cff9daf), torch 2.7.1+cu128, B200
Activity