# Visual Production Reset for Final-Visual Shorts

Use this when a final-near short has repeated visual proof failures: blind T2V was too abstract/generic, keyframes were not I2V-ready, or a fallback animatic was mistaken for progress.

## Hard reset trigger

If the user says the problem is the production method, not just quality, stop generating content clips. Typical signals:

- keyframes are not visually coherent;
- start/end frames have different style, perspective, lighting, or scene layout;
- output is only a morph/overblend, not real video motion;
- Wan/I2V failed technically;
- an animatic was delivered as if it were a proof.

Immediate actions:

1. Reject/audit the current proof.
2. Set `full_final_render_allowed=false` and `basis_for_final_render=false`.
3. Forbid further content visual proofs until technical I2V sanity passes.
4. Run ComfyUI node/workflow audit.
5. Run one minimal technical I2V sanity test with a non-content test image.
6. Only if sanity passes, generate at most 3 styleframes for direction review.

## Proof classification

- `technical_i2v_sanity`: validates the model/workflow; may use trivial test imagery.
- `styleframe_direction_review`: still image only, no Wan video.
- `content_i2v_proof`: short Wan clip from accepted keyframes/style direction.
- `animatic`: internal composition aid only; never a visual-proof substitute.
- `full_final`: requires accepted scene proofs or accepted keyframes plus valid render pipeline.

## Hard rules

```text
prompt_only_generation = true
→ blocks content visual proof

Wan/I2V failure + fallback animatic
→ not a proof, not a basis for final render

full final render
→ requires accepted scene proofs or accepted keyframes
```

T2V-only can be used only for internal technical experiments, not for content proof progression after the user rejects blind T2V.

## Styleframe acceptance rules

Before generating endframes or video, pick one styleframe direction. Each styleframe should obey:

- 9:16;
- clean high-end 3D/CGI business-tech motion design;
- same scene idea across variants;
- clear visual mechanism visible within 1 second;
- no baked text, logos, fake UI labels, or fake document writing;
- no toy/miniature figures unless explicitly requested;
- no photoreal office clutter if it distracts from the mechanism;
- no dark sci-fi tunnel or abstract fantasy form.

For first-last-frame I2V later, start and end must have same camera, perspective, lighting, style, and main objects. Only the state changes.

## Recommended user-facing gating commands

```text
APPROVED_STYLEFRAME_DIRECTION <candidate_id> <variant_id>
REVISE_STYLEFRAME_DIRECTION <candidate_id> <variant_id> "..."
REJECT_STYLEFRAME_DIRECTIONS <candidate_id> "..."
```

Do not proceed to an endframe or content I2V proof until the styleframe direction is approved.