The process is simple on the surface. You upload a photo, illustration or product shot, write one or two lines describing what should happen, choose a model, and press generate. A minute or two later you have a clip. What happens in between is not animation in the traditional sense. The model is not sliding layers around like a GIF editor. It is redrawing every frame, using your photo as an extremely strong reference, and trying to keep the result consistent from frame to frame.
The short answer
That redrawing detail explains almost everything about how the tool behaves in practice. A small movement, a blink or a breath or a slight turn, gives the model very little to reinvent, so the clip stays close to the person or object you photographed. A large movement asks it to invent a lot of new pixels, and that is where faces drift, hands multiply and backgrounds slide. Image-to-video is best understood as a controlled hallucination anchored to a real frame.
What the output actually is
What comes back is a video file, not a loop and not an editable scene. It is short. Expect a few seconds, typically five to ten depending on the model, and plan your action around that rather than around a minute-long sequence.
- Length: a handful of seconds. Write one action that finishes inside the clip.
- Format: an ordinary video file you can download, trim and post.
- Continuity: one continuous take, so the camera is one camera throughout.
- Fidelity: the opening frame usually matches your upload closely and later frames drift with the amount of motion you asked for.
- Audio: some models return silent clips, some can produce audio, so check before you plan a voiceover.
Image-to-video compared with text-to-video
Text-to-video starts from nothing. You describe a scene and the model invents the subject, the framing, the lighting and the wardrobe. That is powerful when you have no reference, and frustrating when you do, because getting a specific face or product back out of a prompt is a lottery you keep re-rolling. Image-to-video removes that lottery: the frame is already decided, so the model's job narrows to keeping it and changing a few things over time. The cost is flexibility, since you cannot ask for a different angle or a scene that is not in the photo.
| Approach | Input | What you control | Best for |
|---|---|---|---|
| Image-to-video | One still image | Motion, camera, pacing | Keeping an approved frame exactly as it is |
| Text-to-video | A written description | Everything | New scenes with no reference image |
| Video-to-video | An existing clip | Style and look | Restyling footage you already shot |
Video-to-video is the third option and the one most often confused with image-to-video. It takes a clip as input and returns a clip, usually restyled. If you already have footage and want it to look like something else, that is the tool you want. If you have one frame and want movement, it is image-to-video.
Try a first clip
Open the image-to-video studio, upload one still and describe a single small action. The credit cost is shown before the run, so an experiment costs one job rather than a subscription.
Start from an example
Where it gets used in practice
Most real use is unglamorous and small in scope. The pattern is always the same: an image already works, and a few seconds of movement adds something a still cannot carry.
- Product shots: a slow rotation or a light sweep so a catalogue image feels like something on a shelf.
- Portraits and family photos: a blink, a breath, a slight head turn, enough to bring a still to life without changing who is in it.
- Ads and social posts: a five second clip from a poster or key art, far cheaper than a shoot.
- Illustration and concept art: a gentle push-in that gives a drawing a sense of depth.
- Storyboards and pitches: motion sketches that communicate a shot before anything is produced.
The honest limits
Faces are the first thing to go wrong. A model that holds a still portrait perfectly will start reinterpreting features the moment you ask for a large movement. Hands are worse: fingers are a classic failure, especially when they grasp something or cross the frame. If a hand is visible, keep the motion somewhere else. Text and logos wobble for a related reason, because the model redraws glyphs instead of copying them, and small text breaks first.
Long actions get cut. A clip is one continuous take of a few seconds, so a sequence with a beginning, a middle and an end will not fit. Ask for one action and let it complete. If you need three beats, make three clips and cut them together afterwards.
Starting without wasting credits
Video jobs are priced per run, so a vague prompt is expensive. Video models cost between 60 and 600 credits: Sora 2 is 60, Kling 90, Veo 3.1 is 300, Wan 2.5 image-to-video is 300 with text-to-video from 320, Seedance is 500, and Wan Animate is 600. The free tier includes 50 credits, and paid plans start at $4.90 a month for 300 credits plus 10 daily. Failed runs on the provider side are refunded in credits, so what a bad attempt really costs you is time.
- 1Pick a photo with one clear subject, a simple background and even light.
- 2Ask for a single small action and a static camera.
- 3If the result drifts, reduce the motion before you rewrite the prompt.
- 4Only move on to a stronger camera move once a subtle version looks right.
When you review the clip, watch it once at normal size before you zoom in. Likeness and believability are judged at a glance, the same way anyone else will see it, and a face can survive close inspection while still reading as wrong in motion. If it looks like the person for the first second, you have a usable clip.
Or start from a sentence
No photo to work from? The text-to-video studio generates a scene from a written description, and every model page lists the modes it powers and its credit cost.
Start from an example
Frequently asked questions
- What is image-to-video AI in one sentence?
- It is a generative model that takes a single still image and produces a short video clip in which part of that image moves, while keeping the rest of the frame close to the photo you uploaded.
- How long are the clips it produces?
- Short. Clips are typically five to ten seconds depending on the model, so write one action that finishes inside that window rather than a sequence that will be cut off.
- Do the clips have sound?
- It depends on the model. Some return silent video and some can generate audio alongside the picture, so check the model before you plan a voiceover or a music edit.
- Do I need a prompt if I already uploaded an image?
- Yes, but a short one. The image already carries the subject, colour and lighting, so the prompt only needs to say what moves, how much, and what the camera does.
- What does one run cost?
- Video runs cost 60 to 600 credits depending on the model. The studio shows the exact cost before each job, and failed runs on the provider side are refunded in credits.
Turn a still into a clip
Open the studio, upload an image, describe the motion in one line and pick a video model. The credit cost is shown before the job runs.