MLT Stock Image Automation Assistant

MLT Logo MLT Stock Image Automation Assistant

Documentation & Usage Guide

MLT Stock Image Assistant is a Windows GUI tool designed to streamline the entire process of AI-based stock image generation, tagging, and publishing to stock sites such as Dreamstime. Built for creators and coders, it combines Ollama, ComfyUI, and Python automation under one user-friendly interface.

🔗 Quick Links

🚀 Usage Sequence & Button Guide

App Screenshot

🧠 Feature Highlights

  • Prompt Generation with Ollama LLM (e.g., Mistral)
    • Automatically creates 10 high-quality prompts using a local language model
    • Clean format with Title, Description, and Keywords (new lines for each)
  • Flattened Prompt Conversion
    • Converts prompts into single-line ComfyUI-compatible format for batch processing
  • One-click ComfyUI Launch
    • Launches ComfyUI and waits until the API is ready
  • Automated Image Generation
    • Queues jobs into ComfyUI using SDXL workflow
    • Randomizes model seed
    • Uses: sd_xl_base_1.0.safetensors, sd_xl_refiner_1.0.safetensors, dpmpp_2m_sde_gpu, karras, RealESRGAN_x4plus.pth
  • Image Completion Wait Logic
    • Waits for expected number of JPGs to be generated before proceeding
  • Metadata Embedding + CSV Creation
    • Injects Title, Description, Keywords into JPG IPTC metadata
    • Generates Adobe Stock-compatible CSV
  • Output Management
    • Sorts JPG and PNG files into timestamped folders (YYYYMMDDHHMM)
    • Ensures no overwrite of previous batches
  • FTP Upload (optional)
    • Uploads latest image batch via FTP to a stock site
    • LATEST = most recent timestamped folder

📘 How to Use + Tips

1. Generate Prompt

  • Enter theme/idea into the “Prompt Theme” box
  • Built-in negative prompts avoid malformed hands, limbs, or NSFW content

2. Open Prompt

  • You may paste prompts from other tools
  • Format must be in 3 lines format without bullet points:
    • Title:
    • Description:
    • Keywords:
  • Paste the prompt twice (PRO) for more image variety
  • This is the source for embedding metadata into JPGs

3. Flatten Prompt

  • Converts prompts to single-line format for ComfyUI
  • No prompt limit in PRO version

4. Generate Image

  • Token errors (>77) in ComfyUI are expected and safe to ignore
  • If a JPG metadata mapping fails, the next ones will still work

5. Open JPG Folder

  • Output folders use timestamp format YYYYMMDDHHMM
  • JPGs are separated and upscaled with RealESRGAN_x4plus.pth to ensure size ≥ 3MB for 1344 x 768 & 768 x 1344.
  • Important: Manually delete any low-quality images before uploading
  • CSV file is generated with Title/Keywords ready for Adobe Stock—even if some images are deleted

6. Upload via FTP

  • Set FTP URL, username, and password in the app
  • Only JPGs in the latest timestamped folder will be uploaded
  • After upload, visit the site to confirm tags/descriptions, then submit

7. Tips

  • If generation is slow, check if Ollama is using GPU—terminate via Task Manager
  • For big batches, restart your PC before starting
  • If restarting midway, exit and relaunch the app to reset queue tracking

8. Known Issues

  • If the last image (e.g., #100) has broken metadata, it might’ve skipped a line
  • Solutions:
  • 🛠 Manual: Use an EXIF viewer (e.g., XnView MP) to check for the skipped image
  • ⚠️ Avoid: Generate smaller batches (e.g., 50 instead of 100)
  • ⚠️ DecompressionBombWarning 
    DecompressionBombWarning: Image size (103219200 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.

    🔍 What It Means:

  • Your image is very large: 103,219,200 pixels (e.g., 12,000 × 8,600).
  • Pillow sets a default limit of 89,478,485 pixels to prevent denial-of-service (DOS) attacks from opening maliciously large files that consume too much memory.