Reverse Prompt Engineer

Reverse-engineer reusable image-generation prompts from a reference image URL or a detailed visual description, powered by a token-protected Codex worker.

This Actor is designed for creators, prompt engineers, AI image teams, and agent workflows that need visual references converted into structured Chinese/English prompts for Midjourney, Flux, Stable Diffusion, Doubao, Jimeng, DALL-E, or model-neutral use.

What It Does

• Accepts a reference image URL or visual description.
• Classifies the image type: logo, typography, landscape, portrait, illustration, 3D/C4D, toy character, or mixed.
• Applies the reverse-prompt-engineer workflow locally through Codex.
• Returns a structured markdown prompt package.
• Labels uncertain inferences instead of inventing hidden details.
• Avoids protected character or living-artist names in generation prompts by converting them into visual traits.

When To Use It

Use this Actor when you need:

• image-to-prompt analysis,
• bilingual Chinese/English prompts,
• Midjourney-style or model-neutral generation prompts,
• negative prompts and recommended parameters,
• an Apify API wrapper around a Codex prompt workflow.

Input

Required:

• One of:
  • referenceImageUrl
  • visualDescription

Optional:

• targetModel: model-neutral, midjourney, stable-diffusion, flux, doubao, jimeng, or dall-e.
• imageType: optional category hint.
• outputLanguage: bilingual, zh, or en.
• includeNegativePrompt: include a negative prompt section.
• includeParameters: include generation parameters when useful.
• notes: extra user intent or constraints.

Example input:

{
  "visualDescription": "Chrome 3D text logo on a blue gradient background, beveled edges, glossy reflection, centered studio composition",
  "targetModel": "midjourney",
  "imageType": "typography-logo",
  "outputLanguage": "bilingual"
}

Output

The default dataset receives one item:

{
  "ok": true,
  "status": "completed",
  "chargedEvent": "reverse-prompt",
  "chargedCount": 1,
  "provider": "codex-local",
  "targetModel": "midjourney",
  "imageType": "typography-logo",
  "outputLanguage": "bilingual",
  "markdown": "## 中文拆解\n..."
}

The markdown field contains:

• Chinese visual breakdown
• Chinese generation prompt
• English prompt
• Negative prompt when useful
• Parameters when useful

Pricing

This Actor uses pay-per-event pricing.

• Event: reverse-prompt
• Price: $0.19 per successful reverse prompt
• Failed validation and worker errors are not charged by this custom event.

Worker Requirement

The Actor calls a protected Codex worker configured by the Actor owner through secret environment variables. Public users do not need to provide worker credentials.

For local development of this project, run:

$npm run serve-apify-worker

Cloud runs require WORKER_URL and WORKER_TOKEN to be configured privately on the Actor. Secret values must never be committed to source control or pasted into issue threads.

Security

• The Actor does not accept shell commands, process settings, or executable paths from user input.
• JSON keys and string values are scanned for script and command-like content.
• Local filesystem paths are removed from public output.
• Worker URL and token are owner-managed secrets, not user input fields.
• The Codex skill runs behind a protected worker, not inside user-controlled request JSON.

Suggested Store Metadata

• Category: AI tools
• Tags: reverse-prompt, image-to-prompt, prompt-engineering, midjourney, flux, stable-diffusion, codex-local, a2a, agent
• Pricing: pay per event, $0.19 per successful reverse-prompt