Auto Draw

What is Auto Draw?

AutoDraw is an artificial intelligence drawing application developed to accelerate visual asset creation for educators, designers, and non-artists. It addresses the common challenge where users lack the manual dexterity or graphics software skills required to construct crisp, recognizable icons using a mouse or trackpad. The system operates by capturing continuous canvas input—stored as coordinate arrays across temporal stroke paths—and passing these geometry vectors to a pattern recognition model. Upon evaluating the visual features, the canvas interface updates a horizontal selection bar with matched vector replacements, which users can convert into clean shapes or export directly to an 800x600 Portable Network Graphics (PNG) file. First, it incorporates an auto-suggestion engine that generates top-candidate iconography within 250 milliseconds of pen-up events. Second, it offers standard object-based editing tools including a freehand fill bucket, geometric shape primitives (circles, rectangles, triangles), multi-font text insertion, and uniform resizing handles. Third, it allows users to export full-canvas or trimmed object compositions as non-lossy graphics for integration into presentations or documentation. Unlike vector design software like Adobe Illustrator, which requires manual path manipulation via Bézier control handles, AutoDraw abstracts complex line path construction into automated pattern matching. While raster paint tools like Microsoft Paint store pixels directly on a static bitmap canvas, AutoDraw maintains an object-oriented spatial model where active strokes remain dynamic vectors until output rendering.

How it works

Under the hood, AutoDraw relies on a Recurrent Neural Network (RNN)—specifically a Long Short-Term Memory (LSTM) architecture trained on millions of vector drawings from the public Quick, Draw! dataset. Rather than analyzing a static two-dimensional image grid like standard Convolutional Neural Networks (CNNs), the model processes the temporal sequence of drawing gestures represented as coordinate offsets and stroke status flags. For instance, when sketching a bicycle, the model registers the initial circle stroke as an ordered coordinate array, followed by stroke offset vectors for the frame tubes. The LSTM updates its hidden state vector at each coordinate point, calculating probability distributions across hundreds of target classes. If the probability score for the bicycle class exceeds a preset confidence threshold (e.g., 0.65), matching icons appear at the top of the interface.

How to use it

1. 1Use the AutoDraw tool.
2. 2Use the AutoDraw tool.
3. 3Use the AutoDraw tool.
4. 4Use the AutoDraw tool.

What it can do

• AI Tools

Use cases

Assumptions and limitations

Understanding the result

Draw and let Google's AI guess what you're drawing. Download as PNG.

Tool details

More image tools

References

• /autodraw — GitHub Repository

  Upstream project · GitHub

Frequently asked