SY - H3 Video v1: MiniMax H3 video with sound, with a speed-up built in
One workflow for MiniMax H3 in ComfyUI: a video with sound from a written story, up to three reference pictures, and optional acting clip, start and end pictures and voice line. The H3 Forecast Speed Up SY node sits in the Render group and starts on mild.
Watch the full test first: four speed-ups side by side, with the sound on.
Stable Yogi Nodes required. Install the pack first (GitHub page). The Model Downloader SY box at the top shows which model files you have and fetches the missing ones after asking.
Measured on my PC (same prompt, same seed)
Woman pouring tea, seed 7: 82 s normal, 58 s with mild.
Clay bowl and cat, seed 7: 86 s normal, 60 s with mild.
Clay bowl and cat, seed 11: 85 s normal, 60 s with mild.
Times are drawing time. The pictures here show the normal clip on top and the mild clip below. With the same seed, mild keeps the scene; small details can change. Set the node to off for the exact normal result.
How to use
At the top: the Model Downloader SY box shows which model files you have.
Group 2: write the story: style, place, what happens, what is said.
Group 4: load one reference picture (groups 5 and 6 add more; they start switched off).
Group 12 Render: keep Speed up on mild, or pick off or strong.
Press Run.
More
All six results with their numbers, plus our other H3 workflows: the full write-up on Stable Yogi. Questions: our Discord.