mmprompt/example_output/README.md
Z User 30cbaeb8f6 Switch plan.json output to the native H3 ref2va prompt format
Each scene prompt is now the six native reference sections —
subject_definitions / summary [reference generation] / retention_analysis /
detailed_description (1-2 style sentences then [Shot 1]) /
overall_soundscape / non_diegetic_music — one line per element, headers on
their own line. The full head-to-toe look (nude-not-bare) lives in
subject_definitions and is restated compressed in the [Shot 1] body; both
cast members are defined per scene against the wired subject references.
verify_plan_rules.py auto-detects the format per shot and validates the
ref2va contract (canonical order, directive, definitions, retention,
style prefix, thin floor), so legacy t2va plans keep passing. Both system
prompts, the assembler prompt and SKILL.md carry the new skeleton,
worked examples and self-checks; example_output/README.md marks the old
run as the t2va craft reference.
2026-10-30 19:28:36 +00:00

49 lines
2.7 KiB
Markdown
Executable File

# example_output — full 12-scene reference run
A complete reference run of the `h3-plan-gate` agent workflow (Step 1 through Step 5)
on a 12-shot story — also a worked demonstration that the workflow scales past the
3-scene example in `h3_agent_workflow_examples/` to 10+ scenes with no code changes.
> **Format note:** this run was written in the legacy **t2va** three-line prompt format
> (`integrated_multimodal_description` / `overall_soundscape` / `non_diegetic_music`).
> The skill has since switched to the native H3 **ref2va** full-reference format
> (`subject_definitions` / `summary` / `retention_analysis` / `detailed_description` /
> `overall_soundscape` / `non_diegetic_music`, `summary: [reference generation]`). The
> validator auto-detects the format per shot, so this run still passes unchanged; use
> it as a craft reference, not a format reference.
## Input
> story: in a functional universe there is a new way of handling criminals when found
> guilty of a crime the criminal gets is sent to a special center that turns the
> criminal in to a permanent diaper dependent toddler, their body goes slowly thru the
> following phases adult ->teen->preteen->child->young child->toddler, 12 shots,
> no dialogue
## What each file is
| File | Stage |
|---|---|
| `outline.json` | Step 1 — the scene guide (PROMPT A): 12 scenes, cast, style, soundscape seed |
| — | Step 2 — the guide was shown to the user and explicitly approved before any scene was written |
| `scene_01.json` … `scene_12.json` | Step 3 — one scene per turn (PROMPT B), composed in story order, each validated with `verify_plan_rules.py fragment ... --prev scene_NN-1.json` before the next was started |
| `plan.json` | Step 4 — `build_plan.py outline.json scene_01.json … scene_12.json -o plan.json`, then `--check`: PLAN OK |
## Run stats
- 12 shots x 10 s, `steps: 20`, no dialogue (every `dialogue_used` is `[]`)
- Description lengths: 222 / 224 / 217 / 203 / 201 / 198 / 191 / 188 / 203 / 199 / 215 / 203 words —
carried by the eight composition elements, not a word band
- Delivered: 2674 frames = **111.42 s** (shot 1 plays full; shots 2-12 each lose 22 frames
to the fixed context overlap trim)
- Transformations between phases are shown as continuous on-screen progressions inside
the step scenes; every description restates `<Subject 1>` with that phase's full
wardrobe because nothing is shared between scenes or from a prefix
## Use
Paste the full contents of `plan.json` into the Production Plan node's `plan_json_input`
and queue once — the node renders each shot as its own clip and chains them through the
H3 context loop into one continuous video. Do not edit `plan.json` by hand: change the
guide or a `scene_NN.json`, re-validate, and re-run the merge.