Upload low-res images, set suffix and AI provider, then process selected rows or batches (200 / 500 / all pending). Each image is sent to OpenAI or Gemini, then saved to the library like the desktop bot.