Download
1 variant available
1.1M
32.1K
165K
MiniMax H3 is licensed by MiniMax under the MiniMax H3 Community License Agreement. That agreement’s Applicable Territory excludes the European Union, the United Kingdom, the Republic of Korea and the United States of America. Your use of H3 and of any H3 derivative is subject to that agreement and its Acceptable Use Policy.
MiniMax H3
Make the next segment continue the shot without replaying its inherited opening. This free package includes two ComfyUI workflows: (The pro version of my models and a lot of workflows and courses are live on my site.)
Normal: 20 sampling steps, no Turbo adapter. This is the configuration demonstrated in the tutorial.
Turbo: 4 sampling steps with the matching reference-model adapter. Use it for faster previews and inspect the result before extending.
Both use native 1536 × 864 at 24 fps. Normal requests 5 + 5 seconds, producing 124 + 124 − 39 = 209 frames / 8.71 seconds. Turbo uses a shorter 5 + 3-second preset: 124 + 73 − 39 = 158 frames / 6.58 seconds. H3 Chain to Video removes the 39 inherited frames from the second opening.
Download the ZIP for both graphs, a fresh fictional-adult SFW sample image, quickstart, test results, attribution and checksums. To switch modes, open the other JSON with its own session name. The continuations differ after testing; the gallery is a demonstration, not a controlled quality benchmark.
Setup
Install ComfyUI 0.34.0+ and H3 Continuous by Greg Henry / Loop Forge. Add the base ref2va model, Qwen text encoder and both VAEs. For Turbo only, install the reference-model 4-step v0.1 adapter from LightX2V / ModelTC. Exact filenames, upstream downloads and folders are in QUICKSTART.md.
What this solves — and what it does not
It removes the inherited replay between segments. The included Turbo preview still drifts sideways and briefly crowds/crops the subject at the left edge. Treat it as a fast draft; Normal is the selected film configuration. This method does not guarantee stable faces, locked framing or clean sound. Inspect every extension before generating more. The package was tested on an RTX PRO 6000 96 GB at 270 W; that is a test setup, not a minimum memory requirement.
The silent gallery compares the two example outputs. The guide explains both presets and includes the tutorial. The free workflows remain useful without a paid course.
Learn the method
Watch the tutorial and read the setup guide
Join the Stable Yogi community through the guide for the workflow resources and further lessons. Please comment with the problem you want tested next: changing faces, camera drift or broken sound. Describe what happens in your clips.
Credits
Example layout, prompts and fresh sample: Stable Yogi. Continuous-overlap node implementation: Greg Henry / Loop Forge (MIT; notice included). Turbo acceleration: LightX2V / ModelTC. Base model and software terms remain applicable. No third-party model weights are bundled.