Skip to content

seed-data.md

Latest commit

 

History

History
37 lines (30 loc) · 1.51 KB

seed-data.md

File metadata and controls

37 lines (30 loc) · 1.51 KB
Raw

Seed data

Two sources, both handled by packages/seed and driven by services/worker CLIs (or small batches from /admin/seed-batches).

Real photos — Unsplash (not_ai)

importUnsplashBatch({ query, count, orientation }):

  • Searches Unsplash (content_filter=high), imports photo metadata, URLs, blur hash, colour, and photographer attribution into media + unsplash_photos.
  • Truth label not_ai, confidence seeded, status approved.
  • Tags: category slug + photorealistic + human-made (spoiler) + image.
  • Triggers the Unsplash download_location endpoint per API guidelines.
pnpm --filter @aiornot/worker seed:unsplash 100 20   # total, per-category

AI variants — OpenAI images (ai)

generateAiVariantsBatch({ count }):

  • Picks recent Unsplash seeds as inspiration parents (source_parent_media_id).
  • Builds a reveal-safe prompt (packages/seed/src/prompts.ts) that keeps the category but changes ≥3 creative details — photorealistic, not a copy.
  • Generates with AI_IMAGE_MODEL (default gpt-image-1), uploads the PNG to R2, stores truth label ai, ai_prompt_summary, ai_model.
  • Requires OPENAI_API_KEY and R2_* storage.
pnpm --filter @aiornot/worker seed:ai-variants 100 10   # total, batch size

Targets

  • Launch: 100 real + 100 AI (200 total).
  • Full: 500 real + 500 AI (1,000 total).

Local demo (no keys)

pnpm seed:demo inserts 12 media with picsum.photos placeholders and simulated crowd stats so the app is fully browsable offline.