通过Fal AI使用OpenAI的GPT Image 2生成或编辑图像——该平台是精细排版领域的佼佼者。

一个Claude Code技能，可以将GPT图片2（ChatGPT Images 2.0）连接到你的代理，只需一个直接插入文件。两个端点：文本转图像和图像编辑（可选遮罩）。同步API，无轮询，单次发布。

为什么选择这个型号

GPT 图片2目前是图像内部可读文本的最佳公开模型。当图片是排版时使用它：

• 黑板、商店招牌、手写海报
• 菜单板、包装样图、杂志封面
• 界面截图/应用商店模型
• 社交卡、横幅、多语言设计

如果是写实肖像或严格保留风格，可以用其他型号（Nano Banana、Flux）。对于文字量大，这款更胜一筹。

快速启动

1.安装技能

将SKILL.md文件复制到你的Claude Code项目的.claude/skills/gpt-image-2/目录中。

2.添加你的Fal AI密钥

去 fal.ai/dashboard/keys 拿钥匙。

3.生成

告诉克劳德密码：

或者补充一下：

两个终点

GPT Image 2 Skill

Generate or edit images with OpenAI’s GPT Image 2 — best-in-class for fine typography — via Fal AI.

A Claude Code skill that wires GPT Image 2 (ChatGPT Images 2.0) into your agent in one drop-in file. Two endpoints: text-to-image AND image edit (with optional masks). Sync API, no polling, single POST.

Why this model

GPT Image 2 is currently the best public model for legible text inside images. Use it when the picture is the typography:

• Chalkboards, store signs, hand-lettered posters
• Menu boards, packaging mockups, magazine covers
• UI screenshot / app store mockups
• Social cards, banners, multilingual designs

For photoreal portraits or strict style preservation, use a different model (Nano Banana, Flux). For everything text-heavy, this wins.

Quick Start

1. Install the skill

Copy the SKILL.md file to your Claude Code project’s .claude/skills/gpt-image-2/ directory.

2. Add your Fal AI key

echo "FAL_KEY=your-fal-ai-key" >> .env

Get a key at fal.ai/dashboard/keys.

3. Generate

Tell Claude Code:

Generate a vintage diner chalkboard reading "TODAY SPECIAL — Lobster Roll $24"

Or for an edit:

Edit https://example.com/photo.png — same scene but everyone is on their phones now

Two endpoints

The skill ships with one combined helper in Node, Python, and bash — pass image_urls to use the edit endpoint, omit for text-to-image.

What You Get

• Sync API — single POST returns the image URL. No polling, no taskId.
• Up to 4 images per call for picking from variants.
• Custom dimensions up to 3840×2160 (aspect ratio ≤ 3:1, multiples of 16).
• Image edit with optional masks for surgical region edits.
• Combined helper in Node, Python, and bash — same shape across languages.

Defaults

Prompting patterns

The community has converged on two patterns the skill documents:

1. Structured JSON prompts for multi-element scenes (posters, infographics, comics, app mockups). Pass the prompt as a JSON object with explicit slots (type, subject, style, background, header, layout). Notably better than prose for complex layouts.
2. Raycast-style placeholders — {argument name="quote" default="Stay hungry, stay foolish"} — for parameterizable prompt templates.