People often write to us about this, and most of it is fixable by you on the next attempt — so here is the honest explanation plus what actually helps.
Why results vary
Evili sends your photo to an AI model, and the model draws a new scene from it. It is not editing your photo and it is not compositing you into a video clip — it is generating, and generation is probabilistic. That means:
- The same photo and the same template can give different results twice. Normal.
- Faces are approximate. A generated face reads as you to a stranger more often than it does to you, because you know your own face better than any model will.
- Text inside an image is often wrong. Letters and words are the weakest part of image generation today. If a template puts writing in the scene, expect it to be imperfect.
- A template that suited one photo may not suit another. The example on the card was made from a different face, in different light.
- A template can change. Templates can change or be withdrawn when the model behind them changes. Your earlier results stay in My generations until their 30 days are up.
No sound, or bars at the edges
- Videos start playing muted. Browsers only let a video play by itself without sound. Press the speaker icon in the top corner of the video to turn the sound on, and tap the video to pause it.
- Bars at the edges. The result screen shows the whole result without cropping it, so a result whose shape differs from the frame has bars around it. Where a template offers Frame, that is the choice that changes the shape.
What actually improves a result
In rough order of how much difference it makes:
- Good light on the face. Even, front-on light. Not backlit, not a dark room.
- The face large and unobstructed. Fill a good part of the frame. No sunglasses, no hand across the cheek, no heavy shadow.
- One person in the photo. If there are several faces, the model may pick the wrong one.
- Front-facing, eyes toward the camera. Sharp profiles and steep angles are harder.
- A real photograph, not a screenshot of a photograph. Screenshots lose detail; so do heavily filtered or already AI-generated images.
- A reasonably large file. A tiny thumbnail has nothing for the model to work with.
If your first attempt was disappointing, changing the photo helps far more than repeating the same one.
Can I get the coins back for a result I don't like?
No, and we would rather explain why than leave it unsaid: the generation ran, and the provider billed us for the compute the moment it did — whether or not you liked the outcome. That is also why the coin cost is shown on the template and on the Generate button before you start.
On a template that makes a frame first, the frame is a finished result too. If you do not like it, Another frame makes a new one and is charged again, and a frame you did not use is not refunded.
A result that came out technically fine but not as you imagined is a normal outcome, not a fault. If you would like another go, Generate again makes a new result and spends coins again.
Tell us about a template that disappoints
If a template keeps giving results that do not look like the example, send us its name at support@evili.ai — a screenshot of the result helps. We look at every report and fix or retire templates that are not working well. This does not refund the coins, but it is how the template gets better for your next try and for everyone else's. A generation that failed and produced nothing is different — those coins come back automatically. See A generation failed, or got stuck.
If a template is plainly broken
If a template produces something obviously wrong every time — the wrong subject, a corrupted image, an output that has nothing to do with the template — that is on us. Write to support@evili.ai with the template name and roughly when you tried it. Your satisfaction is what we are after, so tell us and we will look at the template itself.