Caption Creator is a portable Windows app for generating high-quality text from images. Create captions, tags, JSON, YAML, Illustrious prompts, or fully custom outputs for image datasets, LoRA training, AI prompting, and folder organization.
Run everything locally with built-in GGUF models, or connect your own vision model through LM Studio or Ollama. Your images stay on your computer.
Website · Online Version · Releases · Patreon
- Multiple output types: Captions, Tags, JSON, YAML, Illustrious, and Custom prompts.
- Single or batch workflow: Process one image, many images, or queue multiple jobs.
- Local-first generation: Use bundled models, LM Studio, or Ollama vision models.
- Model management: Download, select, delete, load, and eject models from the app.
- Professional workflow controls: Max words, trigger words, prompt enrichment, Low-VRAM mode, custom output folder, and original filename preservation.
- Fast result actions: Copy output, open the run folder, or export the current run as a ZIP archive.
- Modern desktop UI: Frameless dark interface with live status, progress, previews, gallery output, and an About panel.
| Option | Best for |
|---|---|
| 6GB VRAM (E2B Q4_K_P) | Smaller GPUs and lighter local runs |
| 8GB VRAM (E4B Q4_K_P) | Balanced local captioning and tagging |
| 10GB+ VRAM (E4B Q8_K_P) | Higher-quality local generation |
| 8GB VRAM (NSFW Q4_K_M) | NSFW-focused local generation |
| 12GB VRAM (NSFW Q8_0) | Higher-quality NSFW-focused local generation |
| Custom (LM Studio) | Any compatible local vision model served by LM Studio |
| Custom (Ollama) | Any compatible local vision model served by Ollama |
- Download and unpack the latest release.
- Launch
Caption Creator.exe. - Open Model / VRAM Configuration.
- Download a built-in model that matches your GPU, or
- choose Custom (LM Studio) / Custom (Ollama) and select a running vision model.
- Choose Single Image or Batch Processing, then add images by clicking, dragging, or pasting in Single Image mode.
- Pick an output type: Captions, Tags, JSON, YAML, Illustrious, or Custom.
- Adjust optional settings, then click Generate or add the job to the Queue.
- Copy the result, open the output folder, or save the run as a ZIP archive.
MIT License. See the repository license for details.