Working with language and image models means counting tokens, preparing training data and reproducing results. These utilities cover those chores.
Token counts decide what fits in a context window and what an API call costs. Estimate them before you send a request, then budget with the current price from your provider.
Fine-tuning data must be valid JSONL, and AI-generated images carry prompts and seeds in their metadata. Validate files before uploading a training job, and read metadata to reproduce or learn from a result. Everything runs locally, so prompts and datasets stay with you.
Sketch and iterate on a prompt with the AI prompt builder, count tokens for the final version before sending it, and once you get images or text back, use the metadata reader to inspect exactly how a result was produced.