Pollinations Tarot Deck Generator (Image-first)

Global style (applied to every card)

Keep this consistent to avoid a Frankenstein deck.

Model

Choices come from GET /models if your deployment exposes it.

Width × Height

Use a 2:3 aspect ratio for tarot cards.

Base seed

Each card uses base + index for reproducible variety.

Request delay (ms)

Unauthenticated limit is often ~1 request / 10s per IP. Adjust if you have a token.

Auth token (optional)

If your instance requires a token, it will typically go in the Authorization header.

Safety

Leave on unless you enjoy false positives and moderation ghosts.

Flags

enhance nologo nofeed private

Cache bust (%)

Add random noise to prompts to dodge overly-aggressive caching. 0 keeps prompts clean.

How this builds prompts

Prompt = Tarot card: [Name]. + symbology + your global style. Each card includes canonical elements so the model stops free-styling. Seeds are deterministic per index.

Idle

API Quick Reference (image.pollinations.ai)

GET /prompt/{prompt} → JPEG image. Query: model, width, height, seed, enhance, nologo, nofeed, private, safe.

GET /models → JSON array of available models.

GET /feed → Server-Sent Events for generation events.

Rate limiting often applies to unauthenticated clients. Space requests or use a token.

cURL examples (copy into your terminal)

# 1) Basic
curl -L "https://image.pollinations.ai/prompt/The%20Fool%20tarot?model=flux&width=768&height=1152&seed=1&nologo=true&nofeed=true&safe=true" -o fool.jpg

# 2) With prompt enhancement (if supported)
curl -L "https://image.pollinations.ai/prompt/The%20Magician%20tarot?model=sdxl&enhance=true&seed=2" -o magician.jpg

# 3) Auth header (if your instance requires it)
curl -H "Authorization: Bearer YOUR_TOKEN" \
  -L "https://image.pollinations.ai/prompt/The%20Sun%20tarot?model=flux&width=768&height=1152&seed=19" -o sun.jpg

Troubleshooting

Blank images or repeats: set a base seed and vary by index; keep global style constant; avoid conflicting descriptors.
Cache returning old output: increase Cache bust to 5–15% to append harmless noise tokens.
Rate-limited: raise Request delay to ≥ 11000 ms, or use an authenticated token and lower the delay.
Inconsistent deck look: lock aspect ratio (2:3), fix palette in style, and use the same model for the whole run.
Overcooked compositions: disable enhance and be more explicit with symbology.
Safety false positives: minor body coverage can trigger filters; keep descriptors tasteful or toggle safe=false where allowed.