Ideogram-Json-Captioner Polishes AI Image Datasets With Local Precision

    
        By vramkickedin    
     | 
    
            June 24, 2026 at 8:24 pm        
    
     | 
    
        2 min read

Ideogram-Json-Captioner is a new local desktop application that lets users edit and create structured JSON captions for image datasets. It provides a visual interface for manually editing Ideogram 4 fields like descriptions, styles, backgrounds, and bounding boxes. The tool can also automatically generate captions using local AI models via an OpenAI-compatible server.

Developer Auryg released the tool to help users transfer vetted captions into Ideogram JSON format easily. It runs locally by default and communicates with AI servers like llama.cpp, LM Studio, or Ollama. The app supports manual field editing and can repair or refine auto-generated captions that fail validation.

A visual editor for Ideogram caption workflows

Key features at a glance

Edit all Ideogram JSON fields manually.
Draw, resize, and move bounding boxes on images.
Generate captions with local LLM servers.
Batch auto-caption multiple selected images.
Search captions and filter by keywords.
Retry or clear failed captioning jobs.
Keep original captions separate from JSON output.
Undo the last automatic captioning operation.

This tool is meant for dataset creators and AI image model developers who need captions in Ideogram's structured format. It helps convert plain text captions into JSON and provides a quick way to fix any formatting errors. The batch and search features reduce manual work when preparing large image collections.

Developer tips and known limitations

Auryg advises running a plain text caption first before creating JSON captions for better accuracy. The default prompts sent to the AI are basic, so users should test a few images and adjust prompt overrides if needed. When auto-captioning fails, the app marks images with a failure sidecar and allows retries with a more powerful model or increased context size.