AI Image Generation Automation MVP ## 1. Project Goal Build a scalable AI image generation progran using AdsPower, Playwright, and Supabase. ## 2. Tech Stack & Constraints - **Frontend/UI:** Next.js 14+ (App Router), Tailwind CSS. - **Backend/Database:** Supabase (Auth, Database, Storage). - **Automation:** Python/Node.js Playwright + AdsPower Local API. - **Anti-Detection:** Mandatory use of `playwright-stealth` and `ghost-cursor`. - **Target Platforms:** Google ImageFX, Adobe Firefly, Groq, etc. ## 3. Core Functional Requirements (Must-Have) ### A. Worker & Queue System (The Engine) - **Batch Processing:** Ability to process 400+ prompts in a single session. - **Sequence Indexing:** Every generated file must be named with a leading index (e.g., `001_filename.png`, `002_filename.png`) to maintain order. - **Status Tracking:** Each task must have a status in Supabase: `pending`, `processing`, `completed`, or `failed`. - **Auto-Retry for Failures:** The system must identify `failed` tasks and allow a one-click "Retry All Failures" function to complete the missing indices. ### B. Scalable Plugin Architecture - **JSON-Based Logic:** Do not hard-code site-specific logic. Use a JSON configuration file to store selectors and workflows for each site (ImageFX, Adobe, etc.) so new sites can be added with minimal coding. ### C. Anti-Detection & Stability - **Stealth Mode:** Emulate human behavior (Bezier curve mouse movements, random delays between clicks). - **Session Management:** Use AdsPower's persistent sessions to avoid manual logins for every run. ## 4. Development Priorities (The Golden Rule) 1. **Supabase Schema First:** Set up the `tasks` and `logs` tables to track processing status and indices. 2. **Worker Implementation:** Build the script to read 400 prompts from the DB and save indexed images to local/cloud storage. 3. **Failure Handling:** Implement the logic to re-run only the specific indices that failed. 4. **Minimal UI:** Create a simple dashboard to upload prompts (CSV/Excel) and view the status of each index. ## 5. Strictly Prohibited - **No Image Recognition (OpenCV):** Use only DOM Selectors for reliability. - **No Manual Captcha Solving:** Use persistent manual sessions or 3rd party solvers if absolutely necessary. - **No Synchronous Processing:** All tasks must be processed via an asynchronous Worker-Queue structure.