All artworks below are rendered in real time by the Inkfield engine. Each piece comes with a full stroke-recording JSON that can be replayed directly in the engine, or used as training data for machine learning and style analysis.
Machine Learning Dataset
All recordings are structured JSON containing point-by-point stroke coordinates, millisecond-precision timestamps, strokeData brush parameters, and Flow effect settings. Suitable for generative model training, style transfer research, or painting behavior analysis.
| Artwork | Strokes | Flow | Events | Duration | Size | JSON File |
|---|---|---|---|---|---|---|
| #001 | 8 | 4 | 1,065 | 20.1s | 56.9 KB | Download |
| #002 | 8 | 6 | 938 | 19.1s | 51.9 KB | Download |
| #003 | 8 | 1 | 1,285 | 21.9s | 66.2 KB | Download |
| #004 | 32 | 5 | 1,408 | 25.0s | 99.7 KB | Download |
| #005 | 9 | 2 | 1,487 | 63.5s | 131.1 KB | Download |
| #006 | 17 | 5 | 1,992 | 35.7s | 184.1 KB | Download |
Recording JSON Structure Overview
{
"version": "1.0",
"canvasSize": { "width": 800, "height": 800 },
"canvasBackgroundColor": [222, 212, 195],
"events": [
{ "m": "mp", "t": 150, "x": 400, "y": 300, "strokeData": { ... } },
{ "m": "md", "t": 166, "x": 402, "y": 305 },
...
{ "m": "mr", "t": 1200, "x": 410, "y": 600 },
{ "m": "flow", "t": 2500, "action": "start", "blendType": 3, ... },
{ "m": "flow", "t": 4000, "action": "end", ... }
]
}
For complete field definitions, see
Tutorial #08: AI Agent & JSON Generation;
you can also check the inline API spec in <script id="agent-api-spec"> on the main page.