Skip to main content
The Deltarune Complete Transcript Dataset is a fan-made, structured collection of dialogue, narration, and player choices from Deltarune Chapters 1 through 4. It was created to address a practical gap: large language models trained before or without sufficient coverage of Chapters 3 and 4 lack reliable knowledge of those chapters’ characters, plot, and mechanics. This dataset provides a clean, machine-readable form of the game’s full narrative so that LLMs can be fine-tuned, prompted, or evaluated against accurate Deltarune content.
This dataset is released under the CC0 1.0 Universal public domain dedication. You are free to use, modify, and redistribute it for any purpose without attribution.

​
Data sources

Transcript data was sourced from video playthroughs of each chapter. Raw dialogue and narration were manually transcribed, then segmented into structured records with the assistance of Google Gemini for scene boundary detection and speaker attribution. Each record captures a single utterance or narration block alongside its scene context.

​
Record structure

Every record in the dataset contains exactly three fields:
FieldDescription
contextThe scene or location where the line occurs (e.g., "Scene: Cyber World")
speakerWho is speaking: a character name, Narrator, or Player
textThe raw dialogue, narration text, or player choice option
{"context": "Scene: Device Contact", "speaker": "Narrator", "text": "ARE YOU THERE?"}
{"context": "Scene: Cyber World", "speaker": "Susie", "text": "Hell yeah!!!"}

​
Speaker types

  • Character names (Kris, Susie, Ralsei, Toriel, etc.) — spoken dialogue
  • Narrator — game narration, item descriptions, and visual descriptions
  • Player — selectable choice options presented to the player

​
Available formats

The dataset is distributed in three formats to support different use cases:
  • JSONL — one JSON object per line, one file per chapter (data/chap1_dataset.jsonl through chap4_dataset.jsonl)
  • Parquet — columnar format for efficient querying, one file per chapter plus a combined full_chapters_dataset.parquet (parquet/ directory)
  • ChatML — pre-formatted for LLM fine-tuning (data/chatml/deltarune_story_chatml.jsonl)
Known gaps in the dataset:
  • Chapter 1 was not processed through video-to-text transcription; some scene coverage may differ from later chapters.
  • Chapters 2 and 3 are missing some visual descriptions and environmental narration.
  • The Snowgrave/Weird Route for Chapters 2 and 4 is not included in this dataset. Only the Normal Route is covered for those chapters.

​
Explore the dataset

Quickstart

Load the dataset in Python and run your first queries in minutes.

Schema reference

Full documentation for the context, speaker, and text fields.

Dataset overview

Record counts, chapter breakdowns, speaker distributions, and coverage notes.

LLM fine-tuning

Use the ChatML file to fine-tune a model on Deltarune narrative content.