When a photo-based avatar misses the hairstyle, loses the glasses, or gives you an awkward crop, another click is not always the best next step. First check what the source photo actually shows. This guide focuses on choosing a useful input and changing one variable at a time. For the upload-to-download sequence, see the complete generation walkthrough.
Pick the photograph before writing the prompt
Start with a recent picture in which the person is easy to identify without zooming in. A front-facing or gently angled view is a useful starting point. Keep the hair outline visible, avoid deep shadows over the eyes, and include enough of the clothing to communicate the look you want.
These are practical preparation suggestions, not guarantees about how a model will interpret a face. South Park-style simplification changes proportions, and the result may still omit a distinctive detail.
What a good input keeps, and what gets simplified
The pair below is an unedited first result from our generator on October 17, 2026, with the instruction "Keep the gray beard, brown cap, and green plaid shirt." The input is a synthetic photo made for our demo.
| Synthetic input photo | Actual result |
|---|---|
 |  |
The photo is evenly lit, front-facing, and shows the cap, beard, and shirt clearly, so all three came through. Wrinkles, skin texture, and the exact plaid pattern were simplified into flat shapes, which is how the style works rather than an error. If a feature you care about is missing, check whether it was clearly visible in the input before changing the wording.
| Input choice | What to look for | Useful adjustment |
|---|---|---|
| Lighting | The face and hair are visible rather than silhouetted | Choose a brighter original photo instead of heavily brightening a dark screenshot |
| Framing | The intended person occupies a clear part of the picture | Crop away unrelated subjects while keeping the full hairstyle |
| Accessories | Glasses or a hat are visible and not cut off | Use a source showing the accessory from a simple angle |
| Background | Other faces, signs, and clutter do not compete with the subject | Crop unnecessary background before uploading |
| Image quality | Facial edges remain readable at normal viewing size | Use the original file rather than a repeatedly compressed social-media copy |
Prepare a compatible file
The homepage upload accepts JPG, PNG, and WebP files up to 10 MB. Export an incompatible format to a supported one using your photo editor. Renaming a HEIC file to .jpg does not convert the underlying image.
If the file is too large, reduce its dimensions or export a moderately compressed JPG copy. Keep your original. Do not shrink it so far that the eyes, hair, or accessories disappear into compression artifacts.
For a group photo, crop a copy around the one person you intend to transform. This site's workflow produces a single image per request; it does not automatically detect every person and export a separate avatar for each.
Diagnose the result before regenerating
| What you see | What to check first | One change to try |
|---|---|---|
| The avatar does not feel like the person | Are hair, glasses, and clothing visible in the input? | Choose a clearer photo with the distinguishing feature unobstructed |
| Hair is the wrong shape | Is it blending into the background or cropped off? | Use a better-lit source and briefly describe the hairstyle |
| Glasses or a hat disappear | Is the accessory small, reflective, or partly hidden? | Keep it visible and make it the description's main request |
| Extra people or objects appear | Does the input contain other subjects? | Crop a single-subject copy and simplify the description |
| The head is cut off in a profile circle | Is the face too close to the image edge? | Leave more headroom and preview the final circular crop |
| The result looks soft only after uploading to a profile | Does the original downloaded file look clear? | Start again from the original file and avoid repeated saving or screenshotting |
A visual mismatch and a technical error are different problems. A new prompt will not fix an unsupported file, an exhausted allowance, or a temporary service outage. Follow the message in the tool for those cases.
Use short descriptions to isolate a change
The optional field has a 200-character limit. Rather than rewriting the whole scene after every result, name one important feature: “Keep the round glasses and short black hair” or “Simple background, with space above the head.” These examples express a goal; they have not been benchmarked as guaranteed correction commands.
Keep a note of the source photo, image shape, and wording you used. If you change all three at once, you will not know which change improved the result. Save any version you like before generating another, and account for the cost of each new image.
Know when to switch methods
AI is useful when you want a stylized interpretation of a photograph. It is less suitable when a character must have an exact combination of fixed facial features or clothing parts across many pictures. A manual maker can be a better fit for that task; the manual-versus-AI comparison explains the tradeoff.
If you have no suitable photo, or you want a character who is not a real person, switch the generator to Describe an OC and create one from text alone. The OC guide has tested prompts.
Once the result works, follow the PFP crop checklist instead of regenerating solely to fix a social platform's crop.
