Game Character Sprites for Text-to-Image Generation, 4053 Samples

Saminx22 created a dataset of 4053 captioned pixel-art game character sprites for training image generation models. The collection includes 3445 training and 608 validation samples, with images ranging from 64x64 to 128x64 resolution. Captions were generated using the Qwen2.5-VL-7B-Instruct model, and the dataset was last updated on May 7, 2026.

Use Cases

• Pretraining a text-to-image model for game character design based on the described sprite-caption pairs.
• Fine-tuning a diffusion model to generate pixel-art sprites based on the low-resolution image data.
• Evaluating image generation model performance on a validation split of 608 sprite samples.
• Studying the alignment between text prompts and pixel-art visuals using the provided captions.

Strengths

• Contains 4053 total samples, providing a substantial base for model training.
• Explicitly preserves native low-resolution pixel-art aesthetics by disabling blur and watermark filters.
• Includes a defined validation set of 608 samples for model evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's source and collection method are not described, which may obscure potential biases.
• The license is unknown, which could restrict commercial or research use.

Provenance

Source

huggingface, author Saminx22

Collection Method

Likely collected and captioned from various game asset sources; captions generated by Qwen2.5-VL-7B-Instruct model.

Time Range

null

Freshness

Last updated 2026-05-07 19:03:02; freshness should be verified.

Geography

null