Generate Image

Generate Image

Rules

• Use scripts/generate-image.ts.
• Default endpoint: openai/gpt-image-2. Change --endpoint only if the user asks.
• Run schema before non-trivial payloads.
• Never read or print .env / FAL_KEY values.
• Image-only.

Setup

cp .env.example .env   # from this skill directory
# set FAL_KEY=...

The script loads this skill's .env automatically. Exported env wins.

Workflow

1. auth-check if credentials are uncertain.
2. schema for non-trivial inputs.
3. generate with JSON input.
4. Inspect downloaded files before using them.
5. For official logos/marks: generate scene only, composite real assets locally.

Commands

bun run scripts/generate-image.ts auth-check
bun run scripts/generate-image.ts schema
bun run scripts/generate-image.ts generate \
  --input '{"prompt":"editorial photo of a red fox in Tokyo at night"}'
bun run scripts/generate-image.ts generate \
  --input-file ./input.json \
  --output-dir ./out
bun run scripts/generate-image.ts upload --file ./ref.png

Defaults: queue + poll, download images to a temp dir. Useful flags: --output-dir, --no-download, --logs, --json, --save.

Failures

• 401 / missing key → set FAL_KEY, run auth-check
• 422 → run schema, fix payload
• non-image endpoint → rejected
• COMPLETED + error → fix input or retry