MiniMax Prompt Override

If you build multi-clip MiniMax H3 videos in ComfyUI-Easy-Media, you normally live inside the MultiTrack Editor: click a task segment, type a prompt, set its type and duration, repeat for every shot. That's fine for two clips. It's misery when a shot list is ten segments of "shot → context → context → shot" and you keep re-running with different prompts. MiniMax Prompt Override is the escape hatch: you write your whole per-segment plan as plain text - one prompt per clip, a comma list of durations and modes - and it hands the editor a single override it obeys.

The name undersells the thing slightly. This node doesn't generate anything and doesn't call a model or API. It's a config assembler: it serializes what you type into a JSON payload and emits it from one output socket. The actual magic happens downstream, inside easy multiTrackEditor - wire this node's prompt_override output into the editor's prompt_override input (make sure the editor's format is set to MiniMax) and the editor rebuilds its task, audio, and video tracks from your payload instead of from hand-edited timeline segments. That's the whole deal, and it's why the node is new with v1.3.1: it's the "explicit passing in" path for people driving MultiTrack runs from scripts, agents, or just from a config they want to diff and version instead of clicking.

The inputs that matter

• prompts - one prompt per clip, typed into autogrow rows. This is where your media references live: mention @图片N / @音频N / @视频N or the English @imageN / @audioN / <Picture N> / <Audio N> / <Video N> to pull media from the editor's slot inputs. Only the media types you actually reference get requested, so a text-only clip won't drag a video decode along.
• duration - seconds per clip (2–15). A single 10 applies everywhere; 10,5,10 assigns in order, and missing values reuse the last one.
• generation_type - per clip, comma-separated like everything else: r2v (reference-to-video, the default), i2v (image-to-video), or l2v (last-frame-to-video, the "tail frame" mode).
• continuity_mode - shot generates the segment independently; context inherits the previous segment's motion and audio; context_swap keeps motion while weakening the previous character's look. context_swap is the 1.3.1 addition, so don't assume older tutorial workflows know it.
• system_prompt (optional) - when filled in, it's stamped into every task segment. Leave it empty and each segment keeps whatever default its type implies.
• video_track_lock / audio_track_lock - 1-based index of a video or audio track to lock as the timeline constraint; 0 disables. Locking a video track is the motion-transfer move where the output duration must match the video.

Output is a single prompt_override value, type *, which is the giveaway that it's meant for one specific consumer: the editor's override socket.

Install and gotchas

Install is the pack install: ComfyUI Manager → search ComfyUI-Easy-Media, or cd ComfyUI/custom_nodes && git clone https://github.com/yolain/ComfyUI-Easy-Media, then restart. The pack has zero pip dependencies - but the README is blunt that you want FFmpeg on your system, and actually sampling H3 needs a recent ComfyUI with nested AV-latent support (0.34.0+ for the hard-continuity path). This node itself is featherweight; all the heavy lifting lives in the multitrack project downstream.

Where people will get burned: the override only means something when the editor is in MiniMax format, and the prompts' media references count slots 1-based. Numbers that don't line up with your media list silently produce empty or wrong references. Durations outside 2–15 raise instead of clamping, which is annoying until you remember it. And one real, non-technical catch from the model side: the H3 weights' community licence excludes the US, EU, UK and South Korea, so if you're in those regions the open weights aren't licensed for you to run locally - the API is, the weights aren't. That's a MiniMax problem, not an Easy-Media one, but it'll bite you before this node ever does.