chatgpt-image-ad

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: chatgpt-image-ad
description: >-
  Generate one or more standalone Meta image-ad creatives via ChatGPT Image 2 (gpt-image-2) through the Arcads external API. Locks the model, auto-strips platform chrome, enforces edge-safe layouts and glyph-safety inside body text. Use when the user asks for a "gpt-image-2 ad", "ChatGPT Image ad", "Image 2 ad creative", "make a static image ad with GPT", or anchors on a need for typography-heavy / dense-text / UI-mimicry ad creatives (chat threads, comparison tables, fake search results, iOS dialogs, Slack snapshots, ChatGPT-conversation ads, Apple Notes lists). Does NOT trigger on Nano Banana cues — use nano-banana-image-ad for those.
---

# chatgpt-image-ad (Arcads)

Generate one or more **standalone Meta ad image creatives** via Arcads' `POST /v2/images/generate` with `model: "gpt-image-2"`. Hands the image paths off to your Meta-ad-builder skill — this skill does not upload to Meta itself.

## Read order

1. **This file** — Arcads-specific endpoint, auth, presigned upload flow, workflow phases.
2. **[shared/skills/chatgpt-image-ad/prompting/guide.md](../../shared/skills/chatgpt-image-ad/prompting/guide.md)** — model-specific prompting (what gpt-image-2 is good/bad at, when to switch to nano-banana).
3. **[shared/skills/image-ad-prompting/prompting/prompt-library.md](../../shared/skills/image-ad-prompting/prompting/prompt-library.md)** — 30+ validated templates with per-model notes.
4. **[shared/skills/image-ad-prompting/prompting/safety-suffixes.md](../../shared/skills/image-ad-prompting/prompting/safety-suffixes.md)** — the 3 always-on guards.
5. **[scripts/generate_image.py](scripts/generate_image.py)** — the helper script (Python stdlib only).

## Hard rules — never relax

1. **Model is `gpt-image-2`.** The script refuses any other value. If the user asks for nano-banana, point them at `nano-banana-image-ad`.
2. **No platform/screenshot chrome in output.** `NO_CHROME_SUFFIX` is always on (override with `--allow-chrome` only when the ad's concept *requires* chrome — rare).
3. **Edge-safe + glyph-safety suffixes always on** unless `--no-safe-zone` is explicit. They fix real failures; don't remove silently.
4. **Max 5 reference images.** Hard Arcads cap for gpt-image-2 (observed `400 Max 5 reference image(s) allowed`). The script enforces.
5. **No Meta upload from this skill.** Image generation only. The user has a separate ad-builder skill in their stack — hand off via filesystem paths.
6. **Always present a credit-cost estimate before generating** (see Arcads `arcads-external-api` skill conventions). Each gpt-image-2 call is one image; multiply by `--n` variants.

## Prerequisites

- `.env` containing `ARCADS_BASIC_AUTH` (preferred, pre-encoded) OR `ARCADS_API_KEY`
- Optional: `PRODUCT_ID`, `PROJECT_ID` in `.env` so generated assets land in the right Arcads dashboard folder (see `arcads-external-api` SKILL.md for the session-folder pattern)
- Reference images on local disk (PNG/JPG/JPEG/WEBP/GIF). The script handles the Arcads presigned-upload flow internally — you pass local paths.

## Configuration

- **Base URL:** `https://external-api.arcads.ai` (or `ARCADS_BASE_URL`).
- **Auth:** HTTP Basic. The script prefers a pre-encoded `ARCADS_BASIC_AUTH` env var (e.g. `Basic ZXhhbXBsZTo=`); falls back to encoding `ARCADS_API_KEY` with an empty password.
- **Endpoint:** `POST /v2/images/generate` (request); `GET /v1/assets/{id}` (poll until `status: generated`); image URL fetched once status flips.
- **Reference uploads:** `POST /v1/file-upload/get-presigned-url` returns `{presignedUrl, filePath}`; `PUT` the bytes to `presignedUrl`; pass the `filePath` string in `referenceImages`. **Each filePath is single-use** — the script re-uploads per variant so parallel runs don't collide.

## Generation modes

| Mode | When to use | Required | Optional |
|---|---|---|---|
| `image` (default) | Generate a brand-new ad image. | `--prompt`, `--aspect-ratio` | `--image-ref` (up to 5) |
| `image_edi