MiniMax H3 Motion Repair Plan / 动态风险重做计划 (Advanced)

MiniMaxH3MotionQualityAuditT8Advanced tells you where the footage looks risky; something still has to decide what to do about it. That's MiniMaxH3MotionRepairPlanT8Advanced: it takes the audit's proxy ranges and maps them into the pack's existing non-destructive Selective Repair Plan, so the flagged problem areas become concrete repair jobs with context and seed policy, instead of a list of frame numbers you have to interpret.

The key promises in the description are worth believing: it never renders, overwrites, or automatically accepts media. It produces a plan - a data structure plus JSON - and hands it to whatever will actually execute the repair. If you've been burned by nodes that "helpfully" start generating the moment you connect them, this one is the opposite end of that spectrum.

Inputs

• timeline - an H3_T8_STUDIO_TIMELINE, because the repair plan is anchored to the Studio timeline, not just raw frames.
• audit_report_json - paste the report from the Motion Quality Audit node here.
• audit_scope - single_shot by default, with single_shot_index selecting which shot of the timeline you're mapping.
• mapping_basis - suggested_repair_window by default; this is how audit ranges get turned into repair windows.
• repair_mode - auto by default.
• prompt_addendum - text appended to repair prompts (e.g., "keep the subject's face sharp").
• seed_stride (1009) - how repair seeds step, so each job gets a distinct seed rather than reusing the original.
• context_before_frames / context_after_frames (22 each) - the padding around a flagged range so a repair job has temporal room to work.

Outputs

repair_plan (type H3_T8_REPAIR_PLAN), repair_count (how many jobs the audit produced), repair_plan_json (the machine-readable plan), and mapping_report_json (how the ranges mapped - the audit trail). It's an output node, so it's a natural planning step that ends a graph.

Install and where it fits

Pack install once: ComfyUI Manager → search MiniMax H3 Audio T8, or git clone https://github.com/T8mars/comfyui-minimax-h3-audio-T8 into ComfyUI/custom_nodes/, restart. No pip extras, no model downloads.

Chain position: MotionQualityAudit → MotionRepairPlan → the Studio/repair executor. The workflow is: audit a shot, paste the report JSON in, read repair_count and the plan before running anything. Where people get tripped up: feeding the audit node's report_json instead of its risk_ranges_json, or forgetting the timeline input entirely - this node anchors to a timeline, so it needs one. And keep expectations calibrated: the audit is a proxy, so the plan is only as good as the audit. The pack is explicit that a numeric pass doesn't mean the repair is human-good; the plan just makes the re-render tractable. Experimental, fail-closed, and deliberately non-destructive - a safe place to start if you're curious about the Studio repair system.