Skill-封面

Centrally manages multiple cover styles. Each style has independent dual-aspect-ratio bases and a visual contract; presenter identities and gestures are reused as shared assets.

实战风格

(Real-World Testing Style) is currently registered.

Input Contract

A main title is required. The following optional parameters are accepted:

```
style
```
: Style name or alias; when omitted, the registry's
```
default_style
```
is used.
```
subtitle
```
: Subtitle; when omitted, the subtitle area is left blank, and no conclusions are invented.
```
tag
```
: Top-left label, defaulting to
```
AI 实测
```
.
```
highlight
```
: Phrase to emphasize; this parameter may be passed repeatedly.
```
gesture
```
: Gesture preset, defaulting to
```
zhengzuo
```
(正坐).
```
reference_cover
```
: Any reference cover; when provided, the system enters “replace the person and text from the reference cover” mode.
```
identity_image
```
: Project digital-human/real-person identity frame; required in reference-cover mode.
```
avatar_video
```
: Original digital-human master; used when an identity frame must be captured live.
```
output_dir
```
: A user-specified path takes precedence.

By default, both 3:4 and 4:3 outputs are generated. Reduce the output only when the user explicitly requests a single aspect ratio.

Style and Gesture Routing

Read
```
references/style-registry.json
```
.
Match the style name stated by the user against
```
display_name
```
, slug, and
```
aliases
```
.
If the user does not specify a style, use
```
default_style
```
. If the reference image clearly requires another registered style, prioritize the reference image.
Read the registry entry's
```
reference_doc
```
and load only the detailed visual contract for the selected style.
Read
```
assets/shared/gestures/registry.json
```
and the corresponding gesture's
```
manifest.json
```
; a style references gestures only through
```
default_gesture
```
and
```
allowed_gestures
```
and does not duplicate their definitions.
An unregistered new style or gesture must not be presented as already existing; create its asset package through the separate “new style” and “new gesture” workflows.

Production Workflow

Determine whether to use a registered style or reference-cover person-and-text replacement; select the title, subtitle, tag, and highlighted phrases.
Use the built-in
```
view_image
```
to inspect the base, identity image, gesture image, and style reference returned by the script.

Run

scripts/build_prompt.py

to generate the prompt and absolute reference-image paths for each aspect ratio. For example:

bash

python3 scripts/build_prompt.py \
  --style shizhan \
  --title '史上最强国产开源模型，Kimi K3完全实测！' \
  --subtitle '真实项目全流程挑战｜综合能力超过 Opus' \
  --highlight 'Kimi K3' \
  --highlight '完全实测'

Use the built-in
```
imagegen
```
capability and invoke
```
image_gen
```
separately for each aspect ratio. Mechanically cropping one aspect ratio into another is prohibited.
Pass the reference images strictly in the order returned by the script: the aspect-ratio base controls the overall composition; the shared identity image controls the face, hairstyle, and black T-shirt; the shared gesture image controls only the pose; if the gesture manifest provides an aspect-ratio-specific composition reference, the next image controls the presenter's position, size, and visible area; and the final style reference controls visual calibration. The exact composition rules in the gesture manifest take precedence over the style's general presenter-range rules.
Inspect every output. If any text is incorrect, perform one targeted “text-only correction” edit and explicitly restate that the identity, gesture, composition, background, and colors remain unchanged.
Save to the directory specified by the user. For an existing video project, save to the project's
```
封面/
```
directory; engineering intermediates must not be placed in
```
成片/
```
.
Run
```
scripts/validate_cover.py
```
to check both aspect ratios and basic clarity, then complete a manual visual quality inspection.

Reference Cover Person-and-Text Replacement

When the user asks to “截取我的数字人替换图中人物” (“capture my digital human and replace the person in the image”), “照这张封面改成当前项目” (“adapt this cover to the current project”), or takes a synonymous action:

Prefer capturing a frame from the project's original HeyGen/avatar master rather than the finished video with a circular mask, subtitles, or other UI. Use the finished video only when the original master cannot be found, and record this degradation.

Run the candidate-frame script and use

view_image

to inspect the contact sheet:

bash

python3 scripts/extract_avatar_frames.py <avatar-master.mp4> \
  --out-dir <项目>/封面/avatar-reference

Select a frame with a clear face, complete hair, a natural expression, and no mid-blink state. The identity frame controls only the face, hairstyle, skin tone, and clothing; the reference cover controls only the background, materials, color palette, and editing hierarchy.

Run reference-cover mode:

bash

python3 scripts/build_prompt.py \
  --reference-cover <参考封面.png> \
  --identity-image <选中的数字人帧.png> \
  --title $'深度测评\nQwen3.8 最新实测' \
  --subtitle '三个真实项目｜能玩、能交互、能验收' \
  --highlight 'Qwen3.8' \
  --highlight '最新实测'

Pass the images in the order returned by the script: Image 1 is solely a style/layout reference, while Image 2 is the sole presenter identity reference. Do not inherit Image 1's person, glasses, clothing, title, or pseudo-text.
Invoke
```
image_gen
```
separately for 3:4 and 4:3 and rearrange each layout independently; mechanically cropping the vertical version into the horizontal version is prohibited.
Check presenter identity, all text, short-edge resolution, and exact aspect ratio; save the final images in the project's
```
封面/
```
directory. Keep identity candidate frames in the project archive and out of
```
成片/
```
.

Shared Assets

Black T-shirt identity:

assets/shared/avatar/black-t-identity.png

Gesture master registry:
```
assets/shared/gestures/registry.json
```
Default “正坐” gesture asset package:
```
assets/shared/gestures/zhengzuo/
```
Alternative “摊手” gesture asset package:
```
assets/shared/gestures/tanshou/
```
Gesture specifications:
```
references/gesture-presets.md
```

The identity image controls the face, hairstyle, skin tone, and clothing. Each gesture asset package independently stores reference images, aliases, intended uses, restrictions, aspect-ratio compatibility, and composition rules; a gesture image may control only the pose and must not override the identity image's control of the face. An aspect-ratio composition reference controls only the presenter's position, size, and upper-body visible range and must not copy any old text.

Current Style

shizhan｜实战风格 (Real-World Testing Style)

Style contract:
```
references/style-shizhan.md
```

3:4 base:

assets/styles/shizhan/base/base-clean-3x4.png

4:3 base:

assets/styles/shizhan/base/base-clean-4x3.png

3:4 reference:

assets/styles/shizhan/style-reference/style-reference-3x4.png

4:3 reference:

assets/styles/shizhan/style-reference/style-reference-4x3.png

Adding a Style

When adding

<style-slug>

, complete every step:

Place independent blank 3:4 and 4:3 bases in
```
assets/styles/<style-slug>/base/
```
.

Place two approved style samples in

assets/styles/<style-slug>/style-reference/

Create
```
references/style-<style-slug>.md
```
and record the color palette, composition, typography, presenter placement, and prohibited elements.
Register the display name, aliases, asset paths, aspect-ratio composition, shared identity,
```
default_gesture
```
,
```
allowed_gestures
```
, and core prompt wording in
```
references/style-registry.json
```
.

Test the dual-aspect-ratio prompts with

scripts/build_prompt.py --style <style-slug>

Generate one real sample and pass quality checks for aspect ratio, identity, text, hands, and visual consistency.

Do not copy an entire shared presenter set into every style directory. Override

identity

for a style only when that style genuinely requires a different presenter.

Adding a Gesture

First generate or capture candidate gestures and ask the user to confirm them. After confirmation, create an independent asset package under

assets/shared/gestures/<slug>/

, containing at least

reference.png

and

manifest.json

; when there are significant aspect-ratio differences, also add

reference-3x4.png

and

reference-4x3.png

. Then register the gesture in

assets/shared/gestures/registry.json

and have individual styles reference it through

allowed_gestures

. An unconfirmed gesture must not be marked as

approved

or registered as the default gesture.

Delivery Quality Checks

At minimum, check the following:

The two images have exact 3:4 and 4:3 aspect ratios and use independent layouts.
The selected style is consistent with the registry, style contract, and reference images.
The black T-shirt presenter identity is consistent, with reasonable positioning, scale, and hand structure.
All text is character-for-character correct, with no old title, pseudo-text, or random characters.
3:4 should be 1440×1920 and 4:3 should be 1920×1440; the short edge must not be less than 1000 pixels.
There are no watermarks, platform UI, logos, or additional copy that the user has not authorized.
Reference-cover mode must confirm that Image 1's original person, glasses, clothing, and old title have not carried over.

General prompt rules are documented in

references/prompt-template.md

skill-cover

NPX Install

Tags

SKILL.md Content (Chinese)

Skill-封面

Input Contract

Style and Gesture Routing

Production Workflow

Reference Cover Person-and-Text Replacement

Shared Assets

Current Style

shizhan｜实战风格 (Real-World Testing Style)

Adding a Style

Adding a Gesture

Delivery Quality Checks