How do I turn a photo into a short video with AI?

An image-to-video model takes your photo as frame one, then predicts every frame that follows from your prompt. It never sees anything outside the photo, so whatever appears later is invented: the back of a head that turns, the room behind a door that opens, the side of a product that rotates. Knowing this explains almost every good and bad result.

Three ways to turn photos into video

We ran five kinds of photo through Revid with the motion prompts below, on the Pro tier, at 5 seconds each. Every clip is the first and only take: no retries and no picking the best of several. The notes say what worked and what did not, because the misses teach as much as the hits.

The photo

The video

The photo

The video

The photo

The video

The photo

The video

The photo

The video

The five photos are AI-generated stand-ins, made for this guide so that no real person is animated. The motion works the same way on your own photos.

The photo is most of the prompt. A model cannot add detail that is not there, and every flaw in frame one gets animated too. Check these before you upload:

• Sharp and large enough: at least 1080 pixels on the short side. Blurry photos turn into smeared motion.
• One clear subject: the more people or objects compete, the more the model has to guess.
• Room to move: the subject should not touch the edges. A face cropped at the chin cannot turn its head.
• Even light on the subject: faces in deep shadow are where identity drift starts.
• No text you need to keep: signs, screenshots and labels warp as soon as anything moves.
• No heavy filters or borders: the model animates them as if they were part of the scene.

Image-to-video keeps the framing of the photo. If you ask for a vertical 9:16 video from a landscape photo, the tool has to crop it (you lose the sides) or invent the missing top and bottom (which looks fake). On Revid's Pro tier, image-to-video simply follows the photo's frame.

Crop around the subject with some space above the head and in the direction they look or move. The model needs that space to animate into.

The model already sees the photo. Describing it again ("a woman in a red dress on a beach") wastes the prompt, and can push the model to redraw her. Write only what should change between the first frame and the last one.

Motion prompts for 9 kinds of photo

Copy the one that matches your photo and adjust the details. Each link opens Revid's generator with the prompt filled in on the cheapest tier; upload your photo as the start frame, then generate.

Build a motion prompt for your photo

Three rules for natural motion

• One movement per 5 seconds. A smile and a head turn is enough for a portrait. Asking for more makes faces change.
• Move the camera or the subject, rarely both. When a subject moves a lot, say "the camera stays still". When the camera orbits, keep the subject still.
• Say what must stay the same. "Keep the face unchanged", "keep the label sharp", "keep the watercolor style". Models follow explicit constraints better than implied ones.

Five seconds is the default that works. It is long enough for one natural movement and short enough that the subject still looks like your photo at the end. Go to 8 or 10 seconds for slow, ambient shots like a landscape or a room, and keep people's clips short: every extra second is more invented frames and more drift.

Use an end frame to go from one photo to another

Add a second photo as the end frame and the model animates the transition between the two: a "then and now" of the same person, a before-and-after renovation, a product changing color, a sketch becoming a finished painting. Both photos should have the same framing and aspect ratio, or the transition will warp. On Revid, end frames work on every tier above and on Veo 3.1, but not on Sora 2.

Change one thing per retry. If the face drifts and the background melts, fix the length first: shorter clips solve both more often than not.

A single animated photo is a moment, not a post. On TikTok, Reels and Shorts, what holds attention is a sequence: something to look at, then something new every few seconds, with music or a voice carrying it. A structure that works for photo shorts:

1. 1Open on the most striking photo, animated, with a one-line caption that asks a question or sets up a story.
2. 2Add 3 to 6 more photos, 3 to 5 seconds each, in story order: before, during, after, or then and now.
3. 3Vary the camera moves from photo to photo (push-in, then orbit, then a still shot) so it does not feel repetitive.
4. 4End on the emotional photo, held a little longer.
5. 5Add music that matches the mood, and a voice-over or captions if the photos tell a story.

You can animate each photo in Single media and assemble them in the editor. Or let Revid do the whole thing: the photo to video converter arranges your photos, adds motion, transitions and music, and the memories video maker builds a narrated story around them.

The video at the top of this guide was made exactly this way: the five animated photos from the examples, a seven-line script, an AI voice, captions and music. Revid put it together from those clips for 10 credits, because nothing new had to be generated.

You can try it for free, but every clip costs the provider GPU time, so no tool offers unlimited free photo animation. Free allowances usually come as a few credits, a watermark or a lower resolution.

On Revid, signing up is free, and new accounts usually get free credits to try generation. Animating a photo on the Low tier costs 15 credits for 5 seconds, which makes it the cheapest way to test motion prompts. Regular use is on a paid plan from $39 a month ($32 billed yearly) for 2,000 credits, which is about 130 photo animations of 5 seconds on the Low tier, or 40 on Pro.

Frequently asked questions