Skip to content
Karmyug

Image and caption pairs

Descriptive captions written by native speakers, not machine translated.

15 languagesMultimodal

Overview

Captions written from scratch in each language rather than translated from English, which preserves the objects and relationships a speaker of that language would actually mention first.

Specification

Format
JSONL plus images
Origin
Written natively, not translated

Sample

Illustrative sample. It shows the structure, fields and metadata of this dataset with invented values, not real recordings or records. A real sample pack is sent on request.

Sample records for Image and caption pairs
item_idmodalitiesimage_fileaudio_filetextlanguagealignment_checkqa_status
mm_0001image + textsample_0001.jpg-Customer reports a delayed refund and asks for a status update.Englishtext matches imageaccepted
mm_0002image + textsample_0002.jpg-Agent confirms identity and explains what happens next.Englishtext matches imageaccepted
mm_0003audio + video + textsample_0003.jpgsample_0003.wavUser asks a follow-up that depends on the previous turn.Englishtext matches imageaccepted
mm_0004image + textsample_0004.jpg-Customer switches topic and the agent re-confirms the goal.Englishtext matches imageaccepted
mm_0005image + textsample_0005.jpg-Conversation closes with a summary and a confirmation.Englishtext matches imageaccepted

Quality assurance

  • Native reviewer pass
  • Caption diversity check
  • Object coverage audit

Want a real sample before you commit?

The sample above shows the structure. We also send a representative slice that includes the hard cases, plus the QA report and licence terms. No call required.