# Character three-view reference: separate front, side and back files

A downloadable original adult character set with explicit view roles, a separate first frame and a checklist for consistent clothing and silhouette.

These original SVG/PNG teaching inputs were authored for a method exercise. No video model was run and no generated result is included. PNG files are raster counterparts of the editable SVG sources.

## Input responsibilities

Use actual supported reference slots; front/side/back are separate files. first-frame is a separate image-to-video opening, not an extra promised reference slot.

Use the current controls at /generator/reference-to-video; check actual accepted inputs and settings before submitting.

## Goal

Prepare an appearance brief from three coherent views of one fictional adult. The objective is to make each file’s role reviewable; no upload count, automatic contact-sheet parsing or identity guarantee is implied.

## Preparation

- Download front.png, side.png and back.png as separate images. Their matching teal coat, rust trousers and dark short hair describe one original adult, not three characters.
- Use the front file for visible face and closure, the side file for profile and coat depth, and the back file for rear hair outline and coat seam. Resolve contradictions in replacement assets before submitting.
- Open /generator/reference-to-video and inspect the current upload slots. Use only the number and type of files the current UI supports; if all three cannot be accepted, select the view required by the shot and keep the other views for review.
- The readable diagram is a contact sheet for people, not a claim that a collage is automatically split. first-frame.png is a separate opening-scene input for an image-to-video workflow; do not assume it can be added to the reference workflow simultaneously.

## Copyable brief

```text
Create a new quiet studio scene with the same fictional adult described by the available uploaded appearance references. The front view supplies visible facial features, short dark hair, the front coat closure and rust trousers. If the side view is available, use it for the profile and coat depth; if the back view is available, use it for the rear hair silhouette and coat seam. Use a plain pale-green setting and a fixed medium-wide camera. The adult starts at a slight three-quarter angle, turns the shoulders gently toward the camera and settles. Keep the teal coat, trousers and hairstyle unchanged. Do not create extra people or show a reference sheet inside the scene.
```

## Procedure

1. **Audit the view set** — Compare the front closure, side hem and rear seam before opening a generator. Record that these files depict the same clothing session; an inconsistent input set cannot be repaired by saying “same character.”

2. **Map only the files actually uploaded** — Number the accepted files in the actual UI order and adjust the copyable brief to those files. Delete clauses about an unavailable view; textual references do not upload missing material.

3. **Keep scene composition separate** — Use reference-to-video when requesting a new studio composition from appearance. If the exact supplied scene must be the opening, choose image-to-video and upload the separate first frame according to that mode’s controls.

4. **Review during the shallow turn** — Compare the changing face silhouette, collar overlap and coat side before the adult settles. Use the back image only for visible rear details; a front-facing shot cannot verify every hidden seam.

## Observation record

Use checks.csv to record pass/fail/not-observable and the earliest affected time. Do not pre-fill a pass: there is no generated clip in this package.

- The accepted file order is recorded and matches the brief.
- The video contains one adult rather than three copies from the views.
- The front closure remains on the same coat and does not become a second collar.
- Short hair keeps a compatible outline as the face angle changes.
- Coat hem and trousers remain consistent between front and side visibility.
- No view panel, labels or contact-sheet border appears inside the scene.
- The review distinguishes observed features from back details never visible in the chosen shot.

## Revise one variable

**The generator shows three adults**: Change only the identity-count sentence to one adult whose different views describe the same person; check that a contact-sheet image was not uploaded by mistake.

**The brief refers to files the UI did not accept**: Correct the file-role map to the accepted slots and remove missing-view clauses. Keep the unaccepted files as offline comparison assets.

**The coat gains different seams during the turn**: Replace only the conflicting reference view or reduce the turn so the uncertain region stays hidden, then inspect the same collar-to-hem transition.

## Ending

Save the accepted-file map, prompt and feature observations. Keep the three separate files and the readable sheet together; record any untested hidden detail instead of marking full identity consistency as proven.

## Evidence boundary

Historical parent queries: “prompt for using chracter reference sheet in minimax h3”, “minimax h3 reference images”. Exact long-tail demand is unmeasured. Source: docs/research/pseo-keyword-gap-2026-10-05/evidence/bing-live-keywords.csv. Existing broad tutorial: /blog/character-consistency-reference-video. The owner covers general appearance selection; this package supplies one original character in three independent files and a checklist for view-specific contradictions rather than another general identity guide.
