Sketch-to-image tutorial

How to turn a sketch into AI art

Turn simple shapes into a clearer visual idea. This guide walks through the browser canvas, prompts, styles and output ratios, then shows what to adjust when a result misses the mark.

Open the sketch-to-image tool

1. Sketch a readable composition

Open the AI sketch-to-image canvas and draw one main subject with simple outlines. Start with the largest shapes, then add a few clues such as a window, ears or a horizon. Use clear spacing between objects and leave room at the edges.

You do not need polished line art. A triangle, square and path can describe a mountain, cabin and foreground. Tiny marks and overlapping outlines are harder to interpret. This guide uses the browser drawing canvas, rather than an image-upload workflow.

2. Explain what the shapes represent

Write the subject first, followed by its position, key colors, setting and mood. The sketch supplies visual guidance; the prompt gives meaning to ambiguous shapes. Avoid asking for several competing art styles at once.

A prompt you can adapt

A suggested exercise, not a promise of a particular output.

Sketch: A mountain triangle, a square cabin and a curved path.

A small red cabin at the foot of a tall mountain, a winding path in the foreground, warm yellow window light, early morning mist, soft blue and peach watercolor. Keep the cabin clearly visible as the main subject.

Replace the subject, setting and colors with your own. Words such as “keep” describe your preference; they do not enforce exact pixel positions or guarantee unchanged outlines.

3. Choose a style and how much to expand

Choose an available style, such as Watercolor. Style Strength controls how strongly that style is applied. Creative Expand separately asks the AI to build out the scene with additional detail. It is useful for an imaginative landscape; leave it off when you want a simpler interpretation.

For a baseline, keep Creative Expand off and make one clear style choice. If the picture needs a fuller setting, enable expansion on a later attempt. Neither setting guarantees exact preservation of line work, proportions or object counts.

4. Match the output ratio to your composition

Select an available image model and ratio before generating. A square sketch can produce a different output shape when supported, but the AI may rearrange or extend the composition to fit.

1:1: A centered character or compact scene.
16:9, 4:3 or 3:2: A landscape, wide setting or scene with subjects side by side.
9:16, 3:4 or 2:3: A tall character, rocket or portrait composition.

These are composition suggestions, not a guarantee that every model offers every ratio. Use the choices currently shown in the tool. Keep important subjects away from the edge when experimenting with a new shape.

5. Review, then change one thing

Check the credit cost and generate when the sketch and prompt are ready. Compare the result with your intention: subject identity, composition, color and style. Choose the single most important mismatch before another attempt.

If the cabin is too small, enlarge it in the sketch or put “the cabin fills the foreground” near the start of your prompt. If the colors are wrong, keep the layout unchanged and clarify the palette. If the scene is too busy, turn off Creative Expand. Each generation can vary, so this is an iterative process rather than a deterministic edit.

Save a result you like before continuing. Note the prompt and settings that worked so you have a useful starting point for your next drawing.

Common sketch-to-image problems

The subject became something else: Separate overlapping shapes, make the main subject larger, and name it in the first sentence. Remove background details that could be mistaken for another subject.
The scene has too many invented details: Turn off Creative Expand and describe a simpler background. This asks for less expansion; it does not lock the geometry or prevent all additions.
The style overwhelms the drawing: Lower Style Strength while keeping the same drawing and prompt. The control changes the influence of the selected style, not a guaranteed percentage of line preservation.
The subject is cropped or shifted: Leave more margin around important shapes. Choose a ratio that fits the composition and be prepared for layout changes when moving from a square to a wide or tall output.
Text, hands or repeated objects look wrong: Simplify small details and avoid relying on precise lettering or exact object counts. Add important text afterward in an editor rather than repeatedly generating for spelling.
An option is unavailable or generation fails: Use one of the currently available model and ratio combinations. Read the error before trying again. Failures are not charged; a successful new attempt is charged separately.

Check the cost before each creation

Select an available image model and output ratio, then review the displayed credit cost before generating. Each successful new generation uses credits, including another attempt with a revised prompt. A failed generation is not charged. Available options and prices can change; the tool shows the current choices.

Set a small attempt budget before experimenting. Discuss what you want to change before generating again, rather than submitting the same drawing repeatedly.

Read about credits and pricing