Image Create

image-create

Use this skill for all pure text-to-image generation requests on Starchild.

Covers: logo design, poster design, illustration, meme creation, game assets, social media content, 3D rendering, educational illustrations, fashion design, food photography, pet illustrations, wedding design, holiday marketing, and artistic style creation.

Core principle: call the provided script. Do not re-implement proxy/billing plumbing.

When to use image-create vs image-portrait:

• image-create → user wants to CREATE something from a text description (no face/identity needed)
• image-portrait → user wants a portrait with their face/identity preserved from a reference photo

1. Quick start — basic generation (most common)

⚠️ Execution context — read this first. The code blocks below are Python, not shell commands. Starchild's bash tool runs /bin/bash -c, which cannot parse exec(open(...)) — pasting them directly into a bash command will fail with syntax error near unexpected token 'open'. Also, exec(open(...)) inside python3 -c fails with NameError: __file__ because the script uses __file__ for path resolution.

Use python3 - <<'EOF' with from exports import when calling via the bash tool:

python3 - <<'EOF'
import sys
sys.path.insert(0, "skills/image-create")
from exports import generate_image
result = generate_image(
    prompt="a futuristic city skyline at sunset with flying cars",
)
print(result)
EOF

The heredoc (<<'EOF') preserves all quotes and newlines — no escaping needed.

exec(open('skills/image-create/generate_image.py').read())
result = generate_image(
    prompt="a futuristic city skyline at sunset with flying cars",
)
# result -> {"success": True, "images": [{"local_path": "output/images/..."}], ...}

2. Quick start — with category preset

exec(open('skills/image-create/generate_image.py').read())
result = generate_image(
    prompt="StarChild AI platform",
    category="logo",
    style="tech",
)

3. Quick start — category only (no custom prompt)

exec(open('skills/image-create/generate_image.py').read())
result = generate_image(
    category="3d",
    style="diorama",
)
# Uses the built-in style template as the full prompt

Delivering the result to the user — IMPORTANT

Never hand the user the raw fal.media URL. fal serves files with restrictive CSP headers. The only reliable delivery path is the already-downloaded local file:

1. Use each image's local_path (e.g. output/images/xxx.png) — the script always downloads on success.
2. Tell the user the files are saved to output/images/ and viewable in the workspace file panel.
3. On Web channel, embed inline so the user can preview in chat:

   ![image](output/images/<filename>.png)
   

4. On Telegram / WeChat: send via send_to_telegram(file_path="output/images/...", message_type="image") or send_to_wechat(file_path="output/images/...", message_type="image").

4. Parameters

*At least one of prompt or category must be provided.

Prompt priority: prompt + category/style (enhanced) > prompt only > category + style > category default.

Aspect ratio auto-selection: When not explicitly set, the script picks the best ratio for the category (e.g., 1:1 for logos, 3:4 for posters, 16:9 for banners).

5. Category and style presets

E: Design — Logo (category="logo")

⚠️ AI cannot reliably render text. Generate the icon/symbol only; add text/wordmark in a design tool (Figma, Canva, Illustrator).

Logo prompt anti-patterns (avoid):

• ❌ "logo with text 'Company Name'" — text will be garbled
• ❌ "photorealistic logo" — logos aren't photos
• ❌ "3D rendered logo" — too complex, won't scale down
• ✅ "flat vector logo of [subject], minimal geometric style, single color, white background"

E: Design — Poster (category="poster")

E: Design — Illustration (category="illustration")

E: Design — Meme (category="meme")

K: Game Assets (category="game_asset")

M: Social Media (category="social_media")

N: 3D (category="3d")

P: Education (category="education")

Q: Fashion (category="fashion")

R: Food (category="food")

S: Pet (category="pet")

I: Product Photography (category="product")

Derived from product-photography skill best practices.

Product photography tips:

• Hero shot: product fills 80% of frame, slight 15-30° angle for dimension
• Packshot (Amazon): pure white background, product fills 85%+, no props/text/watermarks
• Always specify lighting: "soft studio lighting", "dramatic rim lighting"
• For e-commerce: "sharp focus", "no shadows" or "subtle shadow only"

Y: Wedding (category="wedding")

Z: Holiday Marketing (category="holiday")

Major holiday calendar for campaign planning:

C: Art Style (category="art_style")

6. Model selection guide

Decision rules:

1. Default: always use nanopro unless the user explicitly requests higher quality.
2. Use gpt when: user says "highest quality", "best quality", "premium", or the design requires precise text rendering (logos with specific text, posters with typography).
3. Use nanopro when: user wants fast results, is iterating on designs, or generating multiple variants.

# Default (fast)
result = generate_image(prompt="cute cat logo", category="logo")

# High quality (user requested)
result = generate_image(prompt="cute cat logo", category="logo", model="gpt")

7. Aspect ratio guide

Platform-specific overrides (auto-applied):

• TikTok cover → 9:16 (1080x1920)
• Instagram story → 9:16 (1080x1920)
• YouTube thumbnail → 16:9 (1280x720)
• Social media banner → 16:9 (1920x1080)
• 小红书 → 3:4 (1080x1440)

The script auto-selects the best ratio when aspect_ratio is not explicitly set.

8. Prompt engineering best practices

The 5-element prompt structure

Every effective image prompt should include:

[subject/content], [style/aesthetic], [composition/layout], [lighting/atmosphere], [quality modifiers]

Key principles

1. Be specific about the subject:

   • ❌ "a logo"
   • ✅ "a minimalist cat silhouette logo for a pet grooming business, clean vector style"

2. Specify the visual style:

   • "flat design", "3D render", "watercolor painting", "photorealistic"
   • "minimalist", "detailed", "abstract", "geometric"

3. Include composition guidance:

   • "centered composition", "rule of thirds", "symmetrical layout"
   • "close-up", "wide shot", "isometric view", "bird's eye view"

4. Lighting matters:

   • Studio: "soft diffused studio lighting", "Rembrandt chiaroscuro"
   • Natural: "golden hour warm light", "dappled sunlight through trees"
   • Dramatic: "dramatic rim lighting", "volumetric light beams", "neon glow"
   • Flat: "even flat lighting with no shadows" (for icons/diagrams)

5. Quality anchors:

   • "professional quality", "print-ready", "4K resolution"
   • "octane render", "Unreal Engine quality", "magazine quality"

6. For text in images (use gpt model):

   • Explicitly state the text: text reading "SALE 50% OFF"
   • Specify font style: "bold sans-serif typography", "elegant script font"
   • GPT model handles text rendering much better than nanopro

Logo-specific prompt tips (from logo-design-guide)

Keywords that work:

flat vector logo, simple minimal icon, single color silhouette,
geometric logo mark, clean lines, negative space design,
line art logo, flat design icon, minimalist symbol

Keywords that fail:

❌ photorealistic logo (contradiction — logos aren't photos)
❌ 3D rendered logo (too complex, won't scale down)
❌ gradient logo (inconsistent results, hard to reproduce)
❌ logo with text "Company Name" (text rendering fails)

Prompt structure for logos:

flat vector logo of [subject], [style], [color constraint], [background], [additional detail]

Children's illustration tips (from book-illustrator)

• Ages 0-2: Simple, bold, high-contrast, clear shapes
• Ages 3-5: Colorful, expressive, engaging characters with movement
• Ages 6-8: More detailed scenes with visual storytelling
• Ages 9-12: Sophisticated illustrations supporting text
• The 3-Color Rule: Limit each character to 3-4 primary colors for visual clarity

Game asset tips (from game-asset-generation)

• Always specify pixel dimensions for sprites: "32x32", "64x64", "128x128"
• For seamless textures: "must tile perfectly with no visible seams when repeated"
• For sprite sheets: specify grid layout "4x2 grid (256x64 total)"
• For icons: "clear silhouette readable at 32x32 pixels"

Example: building a custom prompt

User: "Design a logo for my coffee shop called Bean Dream"

result = generate_image(
    prompt=(
        "flat vector logo of a coffee bean morphing into a crescent moon, "
        "minimalist design, warm brown and cream color palette, "
        "clean lines, white background, "
        "professional branding quality, works at any size"
    ),
    category="logo",
    model="gpt",  # GPT for better detail
)

Example: game character concept

User: "Create a warrior character for my RPG game"

result = generate_image(
    prompt=(
        "female warrior character, ornate golden armor with dragon motifs, "
        "flowing red cape, wielding a glowing enchanted sword, "
        "determined fierce expression, battle-ready stance, "
        "front view T-pose, clean white background"
    ),
    category="game_asset",
    style="character",
)

Example: social media content

User: "Make a TikTok cover about cooking tips"

result = generate_image(
    prompt=(
        "cooking tips video thumbnail, colorful kitchen scene, "
        "fresh ingredients flying in the air, chef's hands visible, "
        "fun energetic vibe, bold visual impact"
    ),
    category="social_media",
    style="tiktok_cover",
    # aspect_ratio auto-set to 9:16
)

9. Intent recognition guide

Use this table to map user requests to the correct category/style/parameters:

10. Using with a reference image (optional)

While this skill is primarily text-to-image, you can provide a reference image for design inspiration:

# Reference image for design guidance
result = generate_image(
    prompt="redesign this logo in a modern minimalist style",
    category="logo",
    image_path="uploads/old_logo.png",
)

# Reference URL
result = generate_image(
    prompt="create a similar style illustration but with a forest theme",
    image_url="https://example.com/reference.jpg",
)

When a reference image is provided, the script uses the /edit endpoint instead of the generate endpoint.

11. Multiple images

# Generate 4 logo variants
result = generate_image(
    prompt="minimalist mountain logo for outdoor brand",
    category="logo",
    count=4,
)
# result["images"] -> list of 4 image dicts

12. Anti-patterns (avoid these)

13. Provided scripts

14. Local testing

Set FAL_KEY env var to call fal.ai directly (bypasses sc-proxy):

# Basic generation
FAL_KEY=your-fal-key python3 skills/image-create/generate_image.py "a cute robot" illustration fantasy 1 nanopro

# Args: <prompt> [category] [style] [count] [model]

15. Troubleshooting

16. Infrastructure (reference)

• Caller → sc-proxy → queue.fal.run/{model} → fal model providers
• All requests must include Authorization: Key fake-falai-key-12345 (proxy injects the real FAL_KEY)
• Pre-charge happens at submit. Poll/result calls are free.
• Final images live at https://*.fal.media/... — public CDN, no auth needed for download.
• Cost tracking via _cost_track.py — records X-Credits-Used from sc-proxy response headers.
• Each skill contains its own _cost_track.py copy (skills are independently deployed).

Model endpoints