AgentSkillsCN

Team Group Photo

当用户要求“拍摄团队合影”、“生成团体肖像”、“制作团队横幅”、“像素风团队图像”、“多人合照”,或需要一张由多位团队成员共同构图的合成像素艺术图像时,应使用此技能。

SKILL.md
--- frontmatter
name: Team Group Photo
description: This skill should be used when the user asks to "create team photo", "generate group portrait", "make team banner", "pixel art team image", "group shot with multiple people", or needs a composite pixel art image featuring multiple team members arranged together.

Team Group Photo

Generate pixel art team group portraits featuring multiple characters arranged in a composition, each maintaining likeness to their individual reference photos.

Key Insight: Transfer, Don't Describe

DO NOT describe faces in your prompt. The more you describe facial features, the more the model will GENERATE new faces instead of TRANSFERRING likeness from the input images.

Instead:

  • Provide all reference images as --input flags
  • Use a simple prompt that instructs TRANSFER of likeness
  • Include the background image as another input

Usage

Use the generate-image skill script directly:

bash
cd ${CLAUDE_PLUGIN_ROOT}/skills/generate-image && bun run scripts/generate.ts \
  "Pixel art team group portrait. Arrange left to right: [Name1], [Name2], [Name3], [etc], [Mascot]. Transfer exact likeness from each input reference. Use the background image exactly. Uniform pixel art style. No text. No border." \
  --input /path/to/person1.png \
  --input /path/to/person2.png \
  --input /path/to/person3.png \
  --input /path/to/mascot.jpg \
  --input /path/to/background.png \
  --aspect 21:9 \
  --size 2K \
  --output /path/to/output.png

Example: B Open Team (Light Mode)

bash
cd ${CLAUDE_PLUGIN_ROOT}/skills/generate-image && bun run scripts/generate.ts \
  "Pixel art team group portrait. Arrange left to right: Kurt, Luke, David, Jason, Dan, Root mascot. Transfer exact likeness from each input reference. Use the watercolor sky background exactly. Uniform pixel art style. No text. No border." \
  --input /path/to/team/kurt.png \
  --input /path/to/team/luke.png \
  --input /path/to/team/david.jpg \
  --input /path/to/team/jason.png \
  --input /path/to/team/dan.png \
  --input /path/to/team/root.jpg \
  --input /path/to/bg-light.png \
  --aspect 21:9 \
  --size 2K \
  --output /path/to/team-group-light.png

Example: B Open Team (Dark Mode)

bash
cd ${CLAUDE_PLUGIN_ROOT}/skills/generate-image && bun run scripts/generate.ts \
  "Pixel art team group portrait. Arrange left to right: Kurt, Luke, David, Jason, Dan, Root mascot. Transfer exact likeness from each input reference. Use the night mountain scene background exactly. Uniform pixel art style. No text. No border." \
  --input /path/to/team/kurt.png \
  --input /path/to/team/luke.png \
  --input /path/to/team/david.jpg \
  --input /path/to/team/jason.png \
  --input /path/to/team/dan.png \
  --input /path/to/team/root.jpg \
  --input /path/to/bg-dark.png \
  --aspect 21:9 \
  --size 2K \
  --output /path/to/team-group-dark.png

What NOT to Do

BAD - Describing faces causes model to generate instead of transfer:

code
## CHARACTER LINEUP (Left to Right)

1. **KURT** (leftmost, front)
   - Bald head
   - Brown/auburn full beard
   - Dark navy suit jacket with tie
   - Friendly smile

2. **LUKE** (slightly behind Kurt)
   - Dark curly/wavy hair
   - Dark beard and goatee
   - Pink button-up shirt

This approach tells the model WHAT to generate, causing it to create generic faces matching the description rather than transferring likeness from the actual input photos.

GOOD - Simple transfer instruction:

code
"Transfer exact likeness from each input reference"

Let the input images speak for themselves. The model will extract the likeness directly from the photos.

Light vs Dark Variants

For theme-aware websites, generate both variants:

  1. Light mode: Use bright/day background image as input
  2. Dark mode: Use dark/night background image as input

Same character references, different background input.

Options Reference

  • --input <path> - Reference image (up to 14 total: 5 humans, 6 objects)
  • --aspect <ratio> - 1:1, 16:9, 9:16, 4:3, 3:4, 21:9
  • --size <1K|2K|4K> - Image resolution
  • --output <path> - Output file path

Troubleshooting

Faces don't match references

  • Remove ALL facial descriptions from your prompt
  • Ensure each reference image is included as --input
  • Use simpler prompt focused on "transfer" language

Background doesn't match reference

  • Include background image as an --input
  • Say "Use the background image exactly" in prompt
  • Don't describe the background, just reference it

Mascot rendered incorrectly

  • Include mascot reference as --input
  • Name the mascot in the arrangement list
  • If still wrong, add brief note: "Root is a yellow frog with banana on head"

Style inconsistent across characters

  • Add "Uniform pixel art style" to prompt
  • Use same reference style for all character inputs
  • Generate at 2K for better detail consistency