Look at any forum where people post their AI-generated profile pictures. Half of them have one of these problems: eyes facing in slightly different directions, teeth that fade into the lips, a chin that looks like it has been smoothed with a melon baller, or earrings that fuse with the side of the face.
The AI is not malicious. It is doing its best with the photo you gave it. The fix is upstream of the filter.
Trick 1: light from the front, not the ceiling
Most AI cartoonizers want a face they can clearly see. Overhead lighting (ceiling lamps, midday sun) leaves shadows under the eyes and chin that the model interprets as features. You end up with a portrait of a person with sunken eyes and a soft jaw.
Stand facing a window during the day. Soft, diffuse light hits your face evenly. Eyes are clear, skin tones are honest, no harsh shadows. This is the single biggest change you can make.
If you cannot get to a window, use any indirect light source - a lamp behind your phone\'s camera, the screen of a second device, a white wall reflecting daylight. The goal is "light coming from the camera\'s direction", not above.
Trick 2: crop tight
Most cartoonizers work on the whole image. If your face takes up 30% of the frame, the model spends 70% of its capacity on the background, your hair, your shirt, the lamp behind you. Plenty of detail gets allocated to things you do not care about.
Crop tight before you apply the filter. Square crop, face filling 70-80% of the frame, top of the head near the top edge, chin near the bottom. This is the framing portrait painters have used for centuries; it is what AI models were mostly trained on.
The result is a sharper, more confident portrait. The model has more capacity to spend on the parts that matter.
Then, and only then, the filter
With a well-lit, well-cropped photo as input, almost any style works. Anime, cartoon, oil painting, pencil sketch. I will use anime as the example because it is the most demanding - the style relies on bold lines and clean shapes, which exaggerate any defect in the source photo.
Pick the anime style. Start the strength slider at 70-75%. Run it.
If the result feels too smooth and "plastic":
- Lower strength to 50%. You will see more of the original photo coming through, which usually adds character.
- Or try a different anime sub-style if your app has them. "Anime sketch" or "manga line" often work better for photos that already have strong contrast.
If the result loses too much of you:
- Crank the strength to 90%. You will get more pure-style and less photographic detail.
- Or pick a model variant focused on portraits rather than general anime.
The face-specific tricks
A few smaller things that move the needle on portraits specifically:
- Keep hair off the forehead. AI models have trouble with bangs and fringes that cross the eyes - the model often blurs the boundary between hair and skin.
- Remove glasses if you can. Or commit to them: tilt your head so the reflections are minimal. AI struggles most when the lens picks up half of a reflection.
- One face per frame. Multi-face photos confuse most cartoonizers. They sometimes blend features from one face onto another.
- Look slightly off-camera. Direct stare into the lens often gives slightly cross-eyed output. A natural sideways glance produces better results.
- Smile gently if at all. Big toothy smiles are hard for AI - teeth and lip boundaries are where the worst artefacts appear.
The "two photos" workflow
For a profile picture I am actually going to use somewhere visible, I take two photos:
- The "for the filter" photo: tight crop, soft front light, neutral expression, no glasses.
- A "for me" photo: looser crop, more natural lighting, normal expression.
The filter version goes through Photo Cartoonizer and becomes the profile picture. The natural version stays in my library as a record. Both serve different purposes; the AI version is for a tile in a chat app, the natural version is what I want to look back at in five years.
Where to back off
Not every photo is filter material. If you have to fight the AI - constantly adjusting strength, switching models, fixing artifacts in an editor afterwards - the source photo was probably not right to begin with. Take a new one.
And if you find yourself caring deeply about the "best" anime version of your face, take a break. The whole point is that it is a small lightweight thing for chat icons. Two minutes of effort, then move on with your day.