[Legacy] H3 Continuum Prompt Plan

H3 Continuum Prompt Plan Preview is a free sanity check: it validates and previews how the pack parses your Fixed / List / Timeline prompt for a multi-chunk run - without sampling anything. In a world where one bad timeline parse means 15 minutes of generation and a melted scene transition, a node that tells you before you queue is worth a lot.

Inputs: prompt_mode (Auto detects Fixed text, ----separated List, or [0-5s] / [Chunk 1] Timeline sections), prompt_script (multiline STRING, default "Prompt"), chunks (3), and chunk_seconds (5.0, with the tooltip noting native H3 is trained for roughly 4–15 second outputs). Outputs: prompt_plan (the validated H3_CONTINUUM_PROMPT_PLAN you can wire into a V1/V2 sampler's prompt_plan input) and report (STRING - the parse result, readable via Core Preview as Text).

This is the "dry run" layer of the pack's prompting system. A Timeline prompt for three 5-second chunks:

[0-5s]
Opening action.

[5-10s]
Continuous middle action.

[10-15s]
Natural ending.

Preview tells you each section maps to exactly one chunk, that your chunk count and section count line up, and that the ranges are sequential - the README is emphatic that matching the section count to chunks matters, because if you produce fewer usable sections the sampler may reuse the previous one. The report will show you exactly where the parser sees a mismatch.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum.git

or via ComfyUI Manager ("H3 Continuum"), then restart. No extra pip packages; needs a MiniMax H3 ComfyUI environment (weights not bundled) and ComfyUI v0.32.0+.

Where it fits now: the current H3 Continuum Sampler V3.3 parses the same Fixed/List/Timeline formats directly and its status output reports the conditioning/parsing result, so a separate preview node is optional on the modern path. But if you're fine-tuning long prompts, or working in a V1/V2 workflow that accepts prompt_plan, this is the cheapest way to iterate on prompt structure. There's no valid reason to skip it when it exists - it runs in milliseconds and catches the mistakes that cost you a full run.