Back to blog

Photo Tips for AI Video: 9 Rules for a Clean Render

Nov 4, 2026FlowVideoGenFlowVideoGen

The reference photo decides how your AI video turns out. Nine practical rules for people, pets and product photos, plus what to do when a render comes out soft.

The reference photo is three quarters of an AI video. The model does not invent your subject, it animates what the photo gives it, so a clean photo renders clean and a compromised photo renders compromise. These nine rules cover the photos that work, the photos that fail, and the fix for each failure.

Rules for photos of people

One person, facing the camera. The model needs the face front and center to lock identity. Profile shots and over-the-shoulder angles leave it guessing, and guesses drift mid-clip.

Even light on the face. Window light or open shade beats direct sun. Harsh side light carves shadows into the face, and the render reads those shadows as features.

Ears and hairline visible. Hats that shade the eyes, heavy fringes, and sunglasses all hide the anchors the model uses to keep a face consistent. Push the hat back or take it off.

No screenshots, no heavy filters. Social media screenshots carry compression that deletes fine detail, and beauty filters reshape the face into something the animation then exaggerates. Original camera files win.

Rules for pets and products

Pet photos: eye level, ears clear. Get down to the animal's height and keep both ears in frame. The Pet Spotlight scene reads markings and ear shape for identity, the same way it reads faces for people.

Product photos: plain background, one light. The orbit in Product Showcase traces the silhouette. A busy shelf behind the product gives the camera move edges to stutter on. A single light source keeps material texture readable through the whole spin.

Fill the frame with the subject. A subject that occupies a small corner of the photo wastes most of the render resolution on empty background. Crop closer before uploading.

The two checks that catch most problems

Before you upload anything, run these two checks. They take ten seconds and prevent most soft renders.

First, look at the photo at full size on your phone and ask: would a stranger recognize this subject from this photo alone? If the answer is no, the render will not save it. Second, check the aspect. A vertical photo for a 9:16 scene, a wide photo for 16:9. When the photo and the render fight over aspect, the model crops, and crops cut faces.

When a render comes out soft, fix the photo first

It is almost never the model. A face that melts mid-clip traces back to a photo with shaded eyes or compression. A product that warps during the orbit traces back to a cluttered background. Swap the reference photo for a cleaner one and run the same scene again before you blame the generator. For how wording affects the render, see the prompt guide, or go straight to the studio and test these rules on your free starter clip.

FAQ

How large should the photo file be?

Anything a modern phone produces is fine, up to the 10 MB upload limit. Resolution matters less than cleanliness: a sharp 1080p photo beats a soft 4K one.

Can I use a photo with more than one person in it?

For single-subject scenes, crop to one person first. The model locks one identity per render, and a second face in frame pulls that lock off target.

Do I own the video made from my photo?

You need the rights to the photo and the consent of anyone in it. Outputs are for your own use, and content rules exclude celebrities and other people's copyrighted material. The terms page has the full policy.