How to Turn a Photo into Video with AI: Kling, Seedance, Hailuo

TL;DR. To turn a photo into a video with AI, upload one sharp frame in image-to-video mode and set 1 or 2 motions with a short prompt: the model animates the picture into a clip of a few seconds. By task: Kling for faces and portraits, Seedance for dynamics and camera moves, Hailuo for everyday scenes and a quick test. The guide covers the step-by-step process, a ready prompt and how to choose a model. All three models are available in the Pixyn Studio.

---

Turning a photo into a video with AI means taking one still picture and getting a short moving clip from it: a light breeze in the hair, blinking, a camera move, clothes stirring. The technology is called image-to-video: the model builds the missing frames between "now" and "a few seconds later", using your photo as the first frame. This guide covers the step-by-step process, a ready prompt and model choice for the task (Kling, Seedance, Hailuo). New accounts get welcome tokens to start.

TL;DR: the short version

• What it is. Image-to-video: upload one photo, and the model animates it into a short clip. With text you set what exactly should move.
• Which model to take. Faces and portraits: Kling (holds features more steadily). Dynamics, action, camera passes: Seedance (ByteDance). Everyday scenes and a price/speed balance: Hailuo.
• What you need. A sharp photo without heavy noise plus a short prompt for 1 or 2 actions. The less you ask for, the cleaner the result.
• Where to do it. In the browser or Telegram via the Pixyn Studio, and you download the result as MP4.
• How long to wait. One clip usually takes 1 to 4 minutes depending on the model and the load.

What image-to-video is and how it differs from text-to-video

There are two different ways to make an AI video, and they are often confused.

Text-to-video: you describe the scene in words and the model draws it from scratch. Full control over the plot, but no control over how exactly the face, the product or the interior will look: it is new every time.

Image-to-video: you give a ready photo as the first frame, and the model animates exactly that. The face stays yours, the product stays yours, the background stays yours. With text you control only the motion: what should move, where the camera goes, what the dynamics are.

For the task "turn a photo into a video" you need the second mode. It also answers the typical requests: animate a portrait, make a model on a product card turn, add motion to the clouds in a landscape, make a "talking" avatar.

If you have no source photo and need to generate a clip from scratch from a description, that is a different task; the AI models catalog helps choose a model for it.

Which model to choose for your task

There is no universal "best" model: each is strong at its own thing. Pick for the specific frame.

Kling: faces, portraits, people

Kling holds facial features more carefully than the others during motion: the eyes do not "float", the expressions look natural, the blinking is realistic. It is the working choice when there is a person in the photo and it matters that they stay recognizable. The calmer the motion in the prompt, the steadier Kling is.

Seedance: dynamics, camera, action

Seedance from ByteDance works noticeably better with camera movement and the overall dynamics of the scene: fly-throughs, push-ins, orbits, flowing fabric, motion in the frame. If the task is to bring a landscape to life, give a dramatic camera pass or show a product in motion, Seedance more often gives a "cinematic" result. Seedance 2.5 also accepts an end frame and makes clips up to 30 seconds.

Hailuo: balance and everyday scenes

Hailuo is a solid all-rounder: it handles everyday scenes, simple motion and light animation of objects well. It is a sensible first attempt when you are not yet sure which kind of motion suits the frame and want to test a hypothesis quickly. Hailuo 02 makes clips of 6 or 10 seconds and also accepts an end frame.

Luma Dream: keyframes and camera control

Luma grew out of 3D tasks, and it shows: Dream Machine accepts a start and an end frame, and the model builds the motion between them. Take Luma when the motion should not be "any", but specific: from point A to point B, with a set camera pass. More in the Luma Dream Machine review.

PixVerse: trending effects and a longer clip

PixVerse has historically been about two things: ready viral effects in one click and an anime look. Version v6 makes clips from 1 to 15 seconds in up to 1080p per pass and can generate sound (music, effects, dialogue), which saves stitching with a voice-over. Take PixVerse when you need a ready social media format fast.

Higgsfield Soul: not animation, but the base frame

Here it is important not to be misled by the name. Soul is a photo model, not a video generator: on its own it does not make a clip. Its job is to assemble a clean, well-crafted frame (light, texture, composition) that then goes as the first frame into a video model. On a good source, Kling and Seedance make noticeably fewer mistakes.

Soul ID is useful separately: you upload several photos of a person, the model remembers the face and holds it on new frames. This solves the classic pain of series: the hero stops "drifting" from frame to frame.

All these models are in the Pixyn Studio: you can run the same photo through different ones and compare which understood the frame best.

Choose by task, not by "rating": portrait, start with Kling; landscape, action or camera, Seedance; an everyday scene or a quick test, Hailuo. If the result does not satisfy you, change the model, not just the prompt.

Model	Strong side	Good for
Kling	Holds facial features steadily	Faces, portraits, people
Seedance	Camera movement and scene dynamics	Landscape, action, camera passes
Hailuo	Everyday scenes, simple motion	Quick test, simple motion
Luma Dream	Start and end frames, a set camera path	Precise motion from A to B
PixVerse	Trending effects, up to 15 s and sound per pass	A fast social media format
Higgsfield Soul	The base frame and holding the face (Soul ID)	Preparing the first frame and series

The table's takeaway: start a portrait with Kling, trust landscape and camera to Seedance, and an everyday scene or a quick test to Hailuo.

How to turn a photo into a video with AI: step by step

Below is a process that works the same in the browser and in Telegram.

1. Prepare the photo. Take a sharp frame without heavy noise, blur or blown highlights. The face or key object should be large and in focus. Ideally horizontal 16:9 or vertical 9:16 for the target platform (Reels, Shorts, a product card).
2. Open the Studio. Go to the Pixyn Studio, choose the video mode and upload the photo as the source frame (image-to-video).
3. Choose a model for the task: Kling for faces, Seedance for dynamics and camera, Hailuo for everyday scenes and a quick test.
4. Describe the motion with a prompt. One or two actions, no more. For example: "a light breeze moves the hair, the person blinks slowly, the camera is almost still". The shorter the list of motions, the fewer artifacts.
5. Set the duration and format. For social media people most often take 5 seconds: a short clip has less time to "fall apart" and looks cleaner. The aspect ratio matches the platform.
6. Run the generation and wait 1 to 4 minutes. If the queue is busy, the time grows.
7. Judge the result critically. Look at hands, fingers, eyes, object edges and text on clothes or packaging: that is where artifacts usually hide.
8. Iterate. Did not like it? First simplify the motion, then change the model, and only then make the prompt more complex. Download the final as MP4.

A ready prompt that works

An image-to-video model does not need the picture described: it already sees it. Describe only the motion: what moves, how the camera moves, the dynamics and the mood. If the result "did not hear" the prompt, repeat the key words more simply or more briefly.

Light natural motion. The person blinks slowly and smiles barely
noticeably. A light breeze moves the hair and the collar. The camera is
almost still, a very slow smooth push-in. Soft daylight,
realistic skin, no sharp movements. Duration 5 seconds.

Why this prompt works:

• One or two micro-motions instead of a long list: the model is not "overloaded" and does not multiply artifacts.
• "The camera is almost still / a very slow push-in": sharp camera moves break the geometry of the face and background more often.
• "No sharp movements" spelled out reduces the risk of "jelly" wobble.
• An explicit duration: short clips are steadier.

For a landscape, replace the body of the prompt with the motion of the environment: "clouds drift, the grass sways, a slow camera orbit from left to right". For a product card: "the model smoothly turns her head to the camera, the fabric sways softly, a slow push-in on the product".

Anti-prompt: what not to ask for in one clip

• Many actions at once: "walks, talks, waves, turns around and takes out a phone". The model will smear everything and break the anatomy. Split it into separate short clips.
• Complex hand work: gestures, typing on a keyboard, sorting objects. The more active the hands, the higher the chance of an "extra finger".
• Sharp text on clothes and packaging: logos and lettering often "float" into an unreadable mess during motion. If text is critical, take a still image or minimal motion.
• A long meaningful scene: image-to-video animates a moment rather than telling a story. For series and sequences you need editing.
• An exact change of angle: "show the same object from behind". The model does not know what is behind and will invent it.

Batch animation: series and product cards

One clip from a photo takes a couple of minutes. But if there are dozens of tasks (animate the whole catalog of product cards, assemble a Reels series from a photoshoot, run a batch of portraits), the manual "upload, wait, download" mode becomes a bottleneck.

For such volumes Pixyn has Workflow, a visual node-based pipeline: you assemble the "photo, animation, platform format" scheme once and then run a batch of frames through it. How it works: the Workflow page.

A particular but common case is animating marketplace product cards: a short motion of the model or product raises engagement in the feed. And if you also need the source photos themselves (for example, an AI photo of a person for an avatar or a catalog), that is an AI photoshoot, whose result is then animated with the same image-to-video.

From an animated frame to Reels

An animated photo is rarely an end in itself: more often it is one piece of a vertical video. A scheme that works without an editing studio:

1. Hook: 1 to 1.5 seconds. Put the animated frame first: motion in the first frame holds attention better than a still.
2. Three short pieces of 3 to 5 seconds. Different angles or different photos, not one clip stretched to a minute.
3. Voice-over separately. Voice and music go on in editing; you do not need to generate them together with the picture.
4. Set vertical 9:16 at generation, not by cropping afterwards: a crop cuts off exactly what you animated the frame for.

For the platforms themselves and their format requirements, it is easier to look at the primary source, for example Instagram's help on Reels.

What it costs in tokens

Prices per generation at the time of publication: Kling: from 15 tokens, Hailuo from 18, Seedance from 20, PixVerse from 20, Luma Dream from 50, Higgsfield from 100. The spread within a model depends on the version, the duration and the resolution.

The studio shows the exact charge before you run it, so check against it rather than the article: the catalog gets updated. Pack contents are on the pricing page.

FAQ

Can I turn a photo into a video with AI for free? New Pixyn accounts get welcome tokens, enough to try animation and see whether the result suits you without paying. After that, it is by plan, see the pricing page.

Which model animates a face in a photo best? For portraits and people, Kling holds facial features most steadily. Seedance is stronger in camera movement and dynamics. Hailuo is a good quick first attempt for everyday scenes. Compare on your own frame.

How many seconds is a video from a photo? Usually 5 to 10 seconds; some models go longer (PixVerse v6 up to 15, Seedance 2.5 up to 30). Short clips (about 5 seconds) look cleaner. For a longer video, segments are stitched in editing.

How do I get fewer artifacts like crooked fingers or a "floating" face? Reduce the number of motions in the prompt, avoid active hand work and sharp turns, generate 2 or 3 takes and choose the best.

Which photo format and quality suit animation? Sharp, in focus, without heavy noise and blown highlights; the key object large. Choose the aspect ratio for the platform: 9:16 for Reels and Shorts, 16:9 for horizontal video, 1:1 or 4:5 for the feed.

Can I animate an old or black-and-white photo? Yes, but it is better to restore it first and colorize it if needed, and only then animate it.

Try it yourself

You can turn a photo into a video with AI right now, and new accounts get welcome tokens.

Create your first video from a photo →: sign-up takes about 30 seconds via email, Telegram or Google, and then the Studio opens with Kling, Seedance and Hailuo.

If the task is large-scale (animating a catalog of cards or assembling a series of videos), set up the visual Workflow once and run it in batches. Questions and bugs go to support: @pixyn_support.