Gangnam Style AI faces don't look like you? How to fix it

• Small faces. A face that's a few pixels wide in a wide shot leaves the model guessing in the close-up.
• Turned heads, sunglasses, masks, heavy filters. The model fills in what it can't see, and fills it in wrong.
• Harsh or dim light. Strong shadows read as different features.
• One reference for three people. When a tool only takes one photo, or mashes three faces into one image, features leak between friends.
• Close-ups. They are where the trend lives and where any mismatch shows. A face that passes in the group shot can fail up close.

1. One photo per person, each facing the camera, chest up.
2. Daylight or even indoor light, no flash glare.
3. No sunglasses, masks, beauty filters or face stickers.
4. A group selfie only if every face is big enough to see clearly. Otherwise ask for separate selfies.
5. The same person in the same spot every time you retry, so you can compare.

We crop in on small faces automatically, and each person's close-up is made from their own photo, never from someone else's. Faces aren't swapped onto a finished video; every frame is made new from your photos, with your own hair and clothes.

AI video still varies. The free preview shows the middle close-up from your photo, so you can tell whether the photo works before paying. The full video is made again from the same photos, so details can differ from the preview. If a video fails, its credits come back; credits for a video that did generate aren't returned.