The photo that works best for an AI video effect is a sharp, front-facing, evenly lit picture where the face is large and the parts of the body the effect needs are in frame: shoulders for a hug, the full body for a dance, the whole animal for a pet. When a video goes wrong, the cause is almost always the photo, and the fix is a different photo, a tighter crop or a better-suited effect. On videosi there is no prompt box to patch it in, which is why picking the right picture matters so much for effects like the AI pet hug.
Below are nine failures that come up again and again, what each one looks like, why it happens and how to fix it. They are drawn from the tips on videosi's effect pages, not from a long list of guesses.
In this guide
- The one rule behind all nine
- 1. The face drifts
- 2. The body is cropped
- 3. Two people at different scales
- 4. A pet's fur changes
- 5. An old scanned print
- 6. Sunglasses, hats and masks
- 7. A busy background
- 8. A group photo
- 9. Low light and backlight
- All nine at a glance
- Choose your photo
- Continue reading
- Questions
What is the one rule behind all nine failures?
An AI video effect can only move what it can see. Anything missing from the photo, such as legs, a hand or the back of a pet, has to be invented, and invented detail is where strange results come from. Anything unclear in the photo, such as a blurry face or a dark coat, is guessed at. So the rule is simple: give the effect a photo that already shows what it needs to move.
That is also why the fix is never a sentence of extra instructions. videosi builds the instructions for each effect in, so the photo, the crop and the choice of effect are the three things you control. Every fix below is one of those three.
1. Why does the face drift?
What you see: the person in the video looks a little like the person in the photo, but not quite. The nose, the eyes or the expression shift as they move.
Why it happens: the face in the photo is too small, out of focus, turned away or partly covered. If the face is smaller than a thumbnail, there is not enough detail for the likeness to hold once the head starts moving. Heavy beauty filters flatten the features and make it worse, since the model starts from a smoothed face.
The fix: use a different photo. Choose the sharpest, most recent, front-facing picture you have, with the face large in the frame and no filter. If the best photo you have is small, pick an effect that moves less, such as the Smile option in old photo animation, which is the gentlest and keeps the likeness best.
2. Why does a cropped body cause odd results?
What you see: legs that bend the wrong way in a dance, arms that appear from nowhere in a hug, hands with the wrong number of fingers.
Why it happens: the effect needs to move a body part that the photo does not show. In AI dance, if the legs are cropped out the model has to invent them. In a hug, a tightly cropped face leaves it to invent the whole body. The same applies to the UNO reverse car clip, which is mostly upper body and one raised arm.
The fix: pick the photo to match the effect.
- Dance: one full-body photo, straight on, with hands and feet in the frame.
- Hug and UNO reverse car: head and shoulders at least, and hands if you have them.
- Only a face photo available: choose an effect that works from a face, such as old photo animation or the anime filter, which redraws a portrait as an image.
3. Why do two people look different sizes in a hug?
What you see: in the AI pet hug the person or the animal looks like a giant next to the other.
Why it happens: a pet hug uses two photos, one of you and one of the animal. If one is a close-up and the other is a full-body shot, the effect has no common scale and the two come out at the wrong sizes.
The fix: match the two photos. Pick pictures with a similar angle and distance, and crop both to the same framing, for example half-body with the shoulders visible. If both people already appear in one photo, upload that photo in both slots instead.
4. Why does a pet's fur change?
What you see: in the AI pet hug the coat changes colour, markings move, or an extra paw appears.
Why it happens: fur, ears and paws are hard for a model, and a pet has markings its owner knows by heart. Two photo problems make it worse. Black pets in dark rooms lose their outline, so the model cannot tell where the coat ends. And a small or blurry pet photo is the usual cause of extra paws or a changed coat.
The fix: use a different photo. Take or find one near a window or outdoors, where the coat holds. Show the whole animal, or a three-quarter view, rather than a face-only close-up. Use the sharpest one you have. For your own photo in the same effect, a kneeling or sitting picture works best, since the hug happens at the pet's height.
5. Why does an old scanned print look wrong?
What you see: the animated face looks glassy, patchy or like someone else, and creases seem to move.
Why it happens: old photo animation works well with black-and-white and faded prints, but two things hurt it. Glare across the face, which is the most common cause of a poor result, and damage such as creases and tears over the eyes or mouth, which the model has to guess at.
The fix: photograph the print properly. Lay it flat in daylight with no flash, shoot from straight above and crop the border. If the family has an undamaged copy, use that one. Keep the movement gentle too: Smile keeps the likeness best, and hands are where older, low-detail photos struggle. Black-and-white prints stay black and white, because the effect does not add colour.
6. What do sunglasses, hats and masks do?
What you see: the eyes or mouth are redrawn, and the person looks different when they turn or lean in.
Why it happens: anything covering the eyes or mouth gives the model less to keep, and it tends to redraw the face. This is most noticeable in clips with a turn to the camera, like the UNO reverse car transition.
The fix: use another photo. Take sunglasses, masks, hats and helmets out of the picture. If you only have a hat photo, crop so the face stays large, or choose an effect where the head barely moves.
7. Why does a busy background ripple?
What you see: floors, mirrors or crowds behind the person wobble while they move.
Why it happens: when a body moves, the model has to decide what happens to everything near it. Walls, parks and beaches hold steady. Crowds, mirrors and patterned floors tend to ripple as the body moves. Loose sleeves and long skirts can blur during fast moves as well.
The fix: use a photo with a plain background, or crop in on the person. For flowing clothes in AI dance, pick Hip-hop, the smoothest of the three styles.
8. Why do faces get worse in a group photo?
What you see: the people at the edges of a group look soft or generic.
Why it happens: the detail available for each face shrinks as the group grows. Small groups of up to four or five keep each face sharp, and in large group shots the faces at the edges soften. Old group portraits can be animated, but each face gets less detail.
The fix: crop to one or two people before you upload, or choose an effect built for groups. The AI Christmas video is made for families and couples, and works best when everyone faces the camera.
9. How does low light or backlight hurt a result?
What you see: the subject is dim or flat, features are missing and the video looks muddy.
Why it happens: a face in shadow against a bright window loses its features. Very dark photos lose the subject in the first second of the Earth zoom out video. Black pets in dark rooms lose their outline.
The fix: use a photo taken with light on the face, from the front or the side. Move to a window or step outside. The exception is the night scenes in the UNO reverse car transition, which are the most forgiving if your photo was taken indoors.
All nine at a glance
| Failure | Fix |
|---|---|
| Face drift | Sharper, front-facing, unfiltered photo |
| Cropped body | Match the framing to the effect |
| Different scales | Two photos at the same distance, or one photo in both slots |
| Pet fur | Whole animal in even daylight |
| Old print | Re-photograph flat, in daylight, no flash |
| Sunglasses and hats | A photo without them |
| Busy background | Plain background or tighter crop |
| Group photo | Crop to one or two people |
| Low light | Light on the face from front or side |
What does a retry cost?
A standard effect costs 10 credits and there are no free generations, so a photo check before you press Generate is worth thirty seconds. The Single pack is $6.99 for 10 credits, which is one video. The Duo Pack is $29 for 50 credits and the Crew Pack is $99 for 250. Credits never expire. If a generation fails, the credits return automatically. See the pricing page.
videosi accepts JPG, PNG, WEBP and HEIC photos up to 10 MB, returns a 9:16 MP4 without a watermark, and deletes photos and results within 30 days. You must be 18 or older, and photos of celebrities are not accepted.
When your photo is ready, open the effect that suits it: AI pet hug for you and an animal, or old photo animation for a portrait from the family box.
Continue reading
- All videosi effects, with the photo needs of each.
- Pet effects.
- Pricing and credit packs.
Questions
What file format and size should I use for an AI video photo?
videosi accepts JPG, PNG, WEBP and HEIC files up to 10 MB, which covers a normal phone photo, including the HEIC files an iPhone saves. Larger files are rejected. The format matters less than the content: a sharp, front-facing picture with a large face will beat a bigger file with a small, blurry one.
Should I edit or filter my photo before uploading it?
No. Heavy beauty filters and colour grading flatten features, and the effect starts from the smoothed face. Upload the original photo instead. On videosi, an unfiltered picture gives a more faithful video, and the only edit worth making is a crop that matches the effect, such as half-body for a hug.
Do I get credits back if the video finishes but looks wrong?
Credits come back automatically when a generation fails, times out or is stopped by a content check. Credits spent on a completed result are not refundable, because the model has already run, unless the file is clearly broken, such as empty or corrupted. That is why checking the photo first matters: a standard effect on videosi costs 10 credits.
Can I fix a bad result on videosi by writing a better prompt?
There is no prompt to write. videosi builds the instructions for each effect in, so the controls you have are the photo, the crop and the effect, plus options such as the hug mood or dance style. If a result is off, try a sharper, better-lit photo before you spend another 10 credits on a second try.
What is the best photo to try first on videosi?
A recent, unfiltered, front-facing photo in even daylight, with the face larger than a thumbnail and the shoulders in frame. It suits most videosi effects, including the AI pet hug, old photo animation and Smile. For a dance, switch to a full-body shot, and for a pet, show the whole animal.