H3 Auto Chain Create Video

MiniMax H3 only makes clips in the 4-15 second range. So if you want a two-minute scene, you chain clips together - and somewhere between the decoder and the stitcher, somebody has to turn raw frames into a proper video object with the right frame rate and audio. That somebody is H3 Auto Chain Create Video.

Despite the name, it doesn't call any API and needs no key. It's pure local packaging: the boring but load-bearing step that makes every clip in the chain actually stitchable.

What it does

This node is part of TBG______'s (Ltamann's) H3 Motion Context Auto-Chain addon - the same author whose Depth Anything V3 fork floats around the ecosystem. You feed it the clip's decoded frame stack, it validates them against what the chain planner expected, matches the audio duration, and hands you a clean VIDEO output at the chain's frame rate.

The mechanism is worth knowing because it explains the one error that actually bites. The node reads clip_frames from the chain_config you connect - the plan produced by the pack's chain-audio node - and compares it to the number of frames in your images. If they don't match, it hard-errors with a message that says to check the H3 Auto Chain Motion Context Trim node. That's the classic beginner trap: you forgot to trim the continuation clips, so clip 2 has extra context frames baked in, and this node catches it before you waste a stitch.

The inputs that matter

Only two required inputs, and you barely touch either:

• images - the decoded, trimmed clip frames (an IMAGE).
• chain_config - from the pack's chain-audio node. The author's own tooltip spells out the gotcha: "This node uses its FPS instead of a separate widget." There is deliberately no FPS widget here. Set FPS upstream in the chain config or your clips won't line up.

Optional stuff you'll mostly leave alone:

• audio - the generated clip audio. If you connect it, the node trims or pads it to exactly match the clip's frame count, so the saved MP4 isn't half a second short or long. Default: no audio.
• bit_depth - 8 or 10. Leave at 8 unless you're deliberately going HDR.
• color_space - sRGB, HDR, or HDR PQ. Again: the HDR options are there for people with an actual HDR pipeline. Don't flip them "to see what happens."

Output is a single VIDEO - wire it into the pack's stitch node (H3 Auto Chain + Stitch for the auto-requeue loop, or H3 Sparse Chain Stitch if you're stitching what's already on disk).

Installing it

This addon targets ComfyUI 0.34.0 and newer - it relies on ComfyUI's native H3 keyframe and reference handling, and explicitly does not patch ComfyUI internals. Install it the normal way:

cd ComfyUI/custom_nodes
git clone https://github.com/Ltamann/ComfyUI-H3-Motion-Context-Auto-Chain-addon

Then restart ComfyUI. It's also findable in ComfyUI Manager by searching "H3 Motion Context Auto-Chain". Dependencies are the usual suspects - imageio-ffmpeg, numpy, Pillow, safetensors - and they're declared in pyproject.toml, so Manager or a manual pip install handles them.

Two things to sort before this node is useful. First, you need a working MiniMax H3 workflow to begin with - that's the ~42.5GB local weights, and note the MiniMax H3 Community License excludes the US, EU, UK and South Korea from running them locally. Second, the README says the original H3 Motion Context package is optional; this addon ships its own standalone implementation and can coexist with the original if you already have it.

Common issues

• Frame-count error on clip 2+ - you missed the Motion Context Trim step. The Create Video node is literally telling you which node to check.
• Clips at the wrong speed or drifting out of sync - you set FPS in a widget that isn't there. FPS comes from chain_config, full stop.
• No audio in the final file - connect the audio input, or make sure the audio path in your H3 workflow actually produced a clip track before this node.

It's the least glamorous node in the pack. It's also the one that turns "a stack of decoded frames" into "a video" - and if you're chaining long H3 scenes, that's the difference between a finished MP4 and a folder of loose frames.