From the creator community
GPT-6 Astra’s Canva Portrait: Why the Small Shapes Work
Zachi’s video of Astra drawing a portrait inside Canva attracted a large audience on X. The striking part is visible in the recording: a photograph becomes a face and shirt assembled from coarse marks and rounded patches of color. The result remains recognizable even though it gives up photographic detail.
The creator later reported about an hour of work and a switch to computer use after a Chrome tool stopped working. For a designer, the useful question is what must survive simplification. This guide explores that question with a Ciyo raster-avatar workflow; it does not claim to automate Canva or deliver the original editable drawing.
Look at the face before counting the shapes
The video shows the portrait being constructed in the Canva interface, with the reference photograph visible. The creator’s replies add process context, including a usage figure tied to their own plan. Neither that usage number nor the reported duration is a general price estimate.
Study the broad silhouette, hairline, face-shadow relationship and shirt shape. Those features carry identity at small sizes. Reproducing every tiny mark would miss the design question the demo makes interesting: how much can the image simplify while still reading as the same person?
Original community post and creator media. Our Ciyo adaptation follows the case study.
Choose a reference that supports simplification
Use your own photograph or a portrait you have permission to adapt. A clear head-and-shoulders image with one dominant light direction is a practical starting point. Avoid choosing a favorite photograph only because its background is beautiful; an avatar will often crop most of that scene away.
Ask Astra to identify visible likeness cues without guessing personality, ethnicity or other traits from appearance. Useful notes describe the contour of the hair, the width of the glasses if present, the angle of the head and the clothing silhouette. Keep the original beside the simplified candidate during review.
| Feature | Preserve | Simplify |
|---|---|---|
| Head silhouette | Relative width, pose and crop | Tiny skin detail |
| Hair and glasses | Main contour and distinctive visible shapes | Individual strands and reflections |
| Face lighting | One consistent light and shadow pattern | Small tonal variations |
| Clothing | Neckline and shoulder shape | Fine fabric texture |
| Background | Enough contrast around the head | Most incidental scenery |
Build one palette before making ten avatars
For our worked direction, use a warm ivory background, three skin-value groups, two dark hair values and a muted blue shirt. These are creative choices, not a claim to describe the source creator’s exact palette. The small number of color roles helps related portraits feel like a set.
In Ciyo, generate one portrait using the approved photograph as reference. Ask for rounded, tile-like painted marks and a clean silhouette. Request a raster illustration explicitly. If your destination requires separately editable vector shapes, plan a distinct tracing or illustration stage in a suitable editor.
Prompt for likeness and style separately
Give the photograph authority over identity and the written brief authority over treatment. That distinction reduces the chance that a strong style reference changes the person’s face. Keep the initial request to one crop so you can judge the result fairly.
Use the prompt below with your own reference, then identify the largest likeness discrepancy before revising. If the eyes drift, asking for a more elaborate background makes the task harder without fixing the important part.
Use my supplied portrait as the identity reference. Create a head-and-shoulders raster illustration built from rounded, tile-like painted marks. Preserve the visible head angle, hair contour, glasses if present, and shoulder silhouette. Use a warm ivory background, a small set of light/mid/dark face values, dark hair values and a muted blue clothing accent. Keep facial landmarks readable at a small avatar size. Do not change age, invent accessories or copy another artist’s exact image. List three visible likeness checks for my review.Judge the avatar in its actual mask
Inspect the candidate as a square and inside the destination’s circular mask. A portrait can look excellent full-size while the circle clips its hair or removes the shoulder shape that balances the composition. Test an actual small display size, such as a 64-pixel preview, as a design exercise rather than a platform specification.
If you need a transparent version, use Ciyo’s Remove Background after approving the portrait, then inspect the hair edge against both light and dark backgrounds. Background removal does not guarantee clean fine edges, so keep the opaque original as well.
| View | Question | Revision |
|---|---|---|
| Full size | Does the person recognize the likeness? | Correct the largest facial or silhouette difference |
| Small square | Do the face and hair remain distinct? | Increase large-value separation |
| Circle mask | Are key contours clipped? | Increase breathing room around the head |
| Light and dark backgrounds | Is the cutout edge clean? | Refine the edge or use the opaque version |
Extend a team set only after one portrait works
An approved example gives the next portrait a style target. Reuse the crop, palette roles, mark size and background treatment while allowing each person’s visible features to remain distinct. A coherent team set should not make everyone look like variations of the same face.
Deliver the original references, accepted raster images and the intended crops. If you publish a process post, show the authorized reference and explain which tool created which artifact. The community demo is a reason to explore a treatment, not permission to reuse its subject or artwork.
Questions about this workflow
Does this recreate editable Canva shapes?
No. Zachi’s source demo shows work inside Canva. The Ciyo adaptation produces a raster portrait, with optional background removal. Editable vector construction is a separate deliverable.
How do I keep the portrait recognizable?
Preserve the head pose, silhouette and major facial relationships before adding detail. Review the image at its intended avatar size and ask the subject to approve the likeness.
Create a portrait that reads at avatar size
Start with an authorized photograph and one simple visual treatment. Approve the likeness before expanding the set.