GPT-Image-2

Generate or edit images through Frevana's AI Factory image API.

This skill returns the validated API response JSON unchanged. Treat data[0].image_url as the primary image URL.

What This Skill Needs

• user-provided prompt or contents
• optional reference image inputs: --image, --image-url, --image-dir, --mask
• optional Frevana image options: n, size, quality, background, outputformat, outputcompression
• FREVANA_TOKEN in the environment variables, or an explicit --token override for the current run
• curl
• bash
• python3

Execution Order

1. Confirm the user provided the image prompt.
2. If the user references one or more local image files, pass each one with --image.
3. If the user references one or more remote image URLs, pass each one with --image-url; the script downloads supported images first, then reuses the existing upload flow.
4. If the user references a local directory of images, pass it with --image-dir; the script recursively collects supported images from that directory.
5. If the user provides a local mask image, pass it with --mask and keep it PNG-only.
6. Do not ask for or pass through image routing overrides; the Frevana script owns them.
7. Prefer the script over ad hoc curl commands.
8. Let the script read FREVANA_TOKEN first.
9. In non-interactive runs, fail fast if the token is missing and tell the user to set FREVANA_TOKEN or pass --token.
10. Return the raw API response JSON, or the first image URL when the user only wants the hosted asset.
11. Save the JSON with --output when useful.

Allowed Options

• --n: 1-10
• --size: auto, 1024x1024, 1536x1024, 1024x1536, 256x256, 512x512, 1792x1024, 1024x1792
• --quality: standard, hd, low, medium, high, auto
• --background: opaque, auto
• --output-format: png, jpeg, webp
• --output-compression: 1-100
• --image: repeatable local file path, .png/.jpg/.jpeg/.webp, each under 50MB
• --image-url: repeatable remote http:// or https:// image URL; downloads supported .png/.jpg/.jpeg/.webp files first, each under 50MB
• --image-dir: repeatable local directory path; recursively loads supported images, up to 16 images total across --image, --image-url, and --image-dir
• --mask: optional local .png path, under 4MB, only when image inputs are present

Command

bash <skill-path>/scripts/generate_image.sh \
  --prompt "A cinematic product photo of a matte black espresso machine on travertine" \
  --size 1536x1024 \
  --quality high \
  --background opaque \
  --output-format png \
  --output ./out/gpt-image-2-result.json

bash <skill-path>/scripts/generate_image.sh \
  --prompt "Turn these product references into one polished hero shot" \
  --image ./refs/front.png \
  --image-url "https://example.com/reference/detail.png" \
  --image-dir ./refs/detail-shots \
  --mask ./refs/mask.png \
  --size 1024x1024 \
  --output ./out/gpt-image-2-edit-result.json

Response Shape

{
  "created": 1777367067,
  "background": "opaque",
  "output_format": "png",
  "quality": "high",
  "size": "1536x1024",
  "data": [
    {
      "revised_prompt": "string",
      "image_url": "https://static.frevana.com/images/user-id/example-image-id.png"
    }
  ],
  "credits_consumed": 1
}

Notes

• Do not pass image routing overrides; the wrapper rejects them.
• Do not pass --background transparent; gpt-image-2 does not currently support transparent backgrounds.
• If the user says "use /path/to/reference.png", map that to --image /path/to/reference.png.
• If the user says "use https://example.com/reference.png", map that to --image-url https://example.com/reference.png.
• If the user says "use the images under /path/to/references", map that to --image-dir /path/to/references.
• Keep prompt/contents required even for image-to-image requests.
• Do not rewrite, proxy, or transform the returned Frevana image URLs unless the user asks for that.