CanvasCraftSFT: Multimodal Tool-Use Trajectories for Image Creation and Editing

CanvasCraftSFT is a supervised fine-tuning subset of the CanvasCraft dataset introduced with the CanvasAgent research. It contains executable multimodal tool-use trajectories designed to teach agents to reason over user requests, call visual tools, and observe intermediate results for complex image tasks. The dataset was created by GML-FMGroup and was last updated on Hugging Face in May 2026.

Use Cases

• Fine-tuning vision-language models for tool orchestration based on the described executable trajectories.
• Training agents on step-by-step reasoning for image editing based on structured tool arguments.
• Benchmarking multimodal AI performance on complex visual task completion based on the provided trajectories.

Strengths

• Dataset is explicitly designed as a supervised fine-tuning subset for a specific research purpose.
• Contains executable trajectories that include structured tool arguments and intermediate visual results.
• Associated with a published research paper ('CanvasAgent: Enabling Complex Image Creation and Editing via Visual Tool Orchestration').

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.

Provenance

Source

GML-FMGroup on Hugging Face.

Collection Method

Created as part of the CanvasAgent research project; likely contains curated tool-use trajectories.

Freshness

Last updated 2026-05-28 11:16:32; freshness should be verified.