I've been chasing consistency in AI video since my first Seedance render. Every test had the same problem: logos would drift between frames. Textures would shift mid-scene. That jacket seam you spent hours modeling? Gone by frame 12.
Then I watched @hellorob's demo on ComfyUI's feed and realized the problem wasn't Seedance 2.0. It was the pipeline.
What the Pipeline Does
This is a ComfyUI workflow running inside a single MCP chat session. You input a product shot — a bottle, a shoe, a jacket — and it kicks out consistent video renders across multiple angles and lighting setups. The secret is the composite control signal, built from three layers:
- DepthAnything V3 — Produces a dense depth map that locks the spatial geometry of every object in the frame. The bottle stays round. The collar stays raised. No warping.
- Canny edge lines — Blended with the depth map to preserve fine detail boundaries. This keeps the edges of logos, stitch lines, and surface textures from bleeding into the background between frames.
- SAM3 mask — Gates the composite to kill the "sprite outline" leakage that plagues Seedance 2.0 renders. That shimmering halo effect around moving objects? Gone.
The composite control signal feeds directly into Seedance 2.0's image-to-video pipeline. One MCP session. One chat thread. Consistent output.
The Exact Technique
Here's the specific approach that made it work: blend DepthAnything V3 with Canny edge lines, gate through a SAM3 mask, and control everything from a single ComfyUI MCP chat session. The result is consistent product shots where tiny debossed logos survive between takes. The whole pipeline is now a reusable Claude skill.
I ran it. It works. This is the kind of pipeline that makes feature-length AI films possible — one character, one lighting setup, one chat session directing every shot.
Want the full ComfyUI workflow? Reply to this email and I'll send you the MCP configuration.