Terraria Gameplay Video Corpus with 841 Hours from YouTube

Infatoshi compiled 841 hours of vanilla Terraria gameplay video, sourced from 310 videos across 141 YouTube channels. The videos were filtered for clean side-view gameplay and uniformly re-encoded to 256x144 resolution at 15 frames per second without audio. This dataset was last updated on June 12, 2026.

Use Cases

• Training video prediction models based on the uniform 256x144, 15 fps video format.
• Learning behavioral priors for agents based on the filtered, side-view gameplay.
• Pretraining world models for simulated environments using the long-form gameplay sequences.
• Benchmarking video generation techniques on a domain-specific, game-focused corpus.

Strengths

• 841 hours of video provides a substantial volume of training data.
• 310 videos sourced from 141 channels suggests diversity in play styles and content.
• Uniform format (256x144, 15 fps, H.264 MP4) simplifies preprocessing pipelines.
• Videos have a minimum duration of 30 minutes, with an average of 2.7 hours, providing long sequences.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect bias inherent to YouTube as a source, such as popularity or creator trends.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

YouTube

Collection Method

Videos were collected from 141 YouTube channels, filtered for specific gameplay criteria, and re-encoded.

Freshness

Last updated 2026-06-12 17:55:17; freshness should be verified.