# Pre-Release Animatic Preview before Wan/I2V

Use this pattern after the user accepts one or more real-life AI styleframe directions but wants to compare the visual/story flow before spending Wan/I2V resources.

## Trigger

The user approves styleframe directions for comparison, e.g. `APPROVED_REAL_LIFE_STYLEFRAME_DIRECTION ... A` and `... C`, but explicitly blocks Wan/I2V/final render and asks for a high-quality storyboard/animatic preview.

## Core rule

A pre-release animatic is not a final animation and not a Wan proof. It is a review artifact to decide:

- which visual language works better;
- whether the scene flow is understandable;
- whether the concept is worth a later Endframe + Wan/I2V proof.

Keep these blocked unless separately approved:

- Wan / I2V / T2V;
- full-final render;
- posting / platform APIs;
- render-series expansion beyond explicitly approved variants.

## Recommended artifact sequence

1. Save a plan/audit with approved variants and blocked actions.
2. Reuse existing voiceover / preview TTS when available.
3. Reuse existing caption segment logic; for this user use static phrase segments with active-word highlight.
4. For each approved style direction, create a small keyframe set:
   - simple scenes: 1–2 images;
   - transformation/mechanism/fix/takeaway scenes: 2–3 images only if needed;
   - no unnecessary image batches.
5. Build deterministic motion only:
   - subtle push-ins;
   - small pans;
   - crossfades;
   - light parallax only if stable.
6. Assemble MP4 with voiceover and captions.
7. Produce per-variant contact sheets and, if useful, QA frame sheets sampled from the rendered MP4.
8. Deliver the MP4s and sheets with verified metadata.

## Five-scene mapping for the decision-target meeting-summary pilot

For `ai-output-autopsy-decision-summary-38s`, the working scene structure was:

```text
s01_fail        visible fail / missing decision target
s02_bad_output  polished but generic summary
s03_mechanism   information compresses while decision target stays separate
s04_fix         evidence gets sorted into blockers / owners / risks / next steps
s05_takeaway    ask for decision evidence
```

Do not treat this exact topic as a universal template; use it as an example of how to preserve the candidate’s approved story beats.

## Image rules

Generated keyframes must remain text-free:

- no readable text;
- no letters/numbers/pseudo-writing on cards;
- no logos or brands;
- no fake UI;
- no people, faces, or hands when the style direction forbids them;
- no baked captions or labels;
- no abstract fantasy shapes that undermine the real-life/work-scene anchor.

Renderer owns all truth-bearing elements:

- captions;
- labels;
- arrows;
- stamps;
- mechanism callouts;
- final takeaway text.

## Caption requirement for this user

Use the corrected default from `static-active-word-captions-and-final-pilot-render.md`:

```text
static phrase segment + active-word highlight
```

Do not regress to one-word-only captions or a sliding word window.

## Delivery metadata checklist

For every MP4, contact sheet, QA sheet, and ZIP/keyframe bundle, report:

```text
file_path
file_exists
file_size
sha256
media_type
duration
resolution/fps for MP4
telegram_delivery_attempted
telegram_delivery_verified
telegram_message_id
```

If media was only created locally, say exactly that rather than implying attachment.

## Comparison guidance

When comparing variants, separate:

- visual impact / retention pull;
- production safety for later I2V;
- busyness / pseudo-writing risk;
- caption-safe area;
- mechanism clarity;
- real-life work-scene credibility.

A stronger social variant may still require a cleanup instruction before Endframe/Wan, e.g. fewer card fragments, no pseudo-line marks, and preserved caption safe area.

## Pitfalls

- Do not call a pre-release animatic a Wan/I2V proof.
- Do not generate a final render after the user approves only preview comparison.
- Do not create 20 image variants; keep the number bound to the approved directions.
- Do not let generated images invent explanatory text or labels. Add them in the renderer.
- Do not omit QA frames: Telegram video thumbnails can hide caption or pseudo-writing problems.