Upload one photo, describe the motion you want, and get back a short video clip with camera movement, subject movement or both. Built on real image-to-video models like Kling v2.6, Wan 2.7 I2V, Ray 3.2 and Hailuo 02, so the motion looks like footage rather than a slideshow.
A photo stops at the moment the shutter clicked. Video keeps going, and for years the only way to get from one to the other was to have filmed the moment in the first place. Image-to-video models close that gap after the fact: you give them a single photo and a sentence describing what should happen next, and a few seconds later you have a clip with a moving camera, a moving subject, or both, built from a picture that never moved at all.
Most photos you already own were never shot as video, and reshooting is rarely an option: the wedding is over, the product sample went back, the grandparent in the photo is gone. Image-to-video generation works from what you have. It reads the picture as a starting frame and extends it forward in time, following whatever motion your prompt describes.
Picasso IA runs this through several models built specifically for the job, including kwaivgi/kling-v2.6, wan-video/wan-2.7-i2v, luma/ray-3.2 and minimax/hailuo-02, all reachable from the same Toolkit without installing anything. Each handles motion a little differently, so the same photo can come back as a slow cinematic push from one model and a livelier subject animation from another; the full catalog is worth browsing if you want to see what else generates video. New accounts start with free credits, enough to test a handful of photos before deciding whether the result is worth paying for.
The single biggest decision is not which model to pick, it is what kind of motion you are asking for. A prompt that names the wrong kind of movement for the photo is the most common reason a first attempt looks stiff or fake, more than any setting in the model itself.
| Motion style | What to type | Best for |
|---|---|---|
| Ambient cinemagraph | Only the water ripples and the leaves sway, everything else stays still | Landscape and nature photos, background loops |
| Camera push or pan | Slow cinematic dolly forward, camera drifts left to right | Product shots, real estate photos, establishing shots |
| Subject motion | She turns her head and smiles, gentle breeze moves her hair | Portraits, headshots, family photos |
| Combined scene motion | Camera slowly orbits as the dancer spins and the crowd cheers | Social clips, event recaps, ads |
| First-to-last-frame bridge | The flower opens fully, petals unfurl one by one | Before-and-after reveals, product transformations |
Naming the style from this table in your prompt, rather than describing the whole scene from scratch, is usually the fastest way to land close to what you pictured on the first or second try.
Camera motion and subject motion also carry different risk. Camera motion keeps the subject still and moves the viewpoint, the safer choice when the subject is a face, since the face itself is never asked to change. Subject motion animates what is inside the frame, which is riskier because faces and hands are where these models most often introduce a distortion. Combining both is the most cinematic option and the most likely to need a second attempt, so generate a subject-only version first to confirm the motion looks right before layering camera movement on top.
The workflow below assumes nothing more than one photo and an idea of the motion you want. It runs the same way whether the source is a product shot, a portrait, or an old family photo you are bringing back to life.
A photo-to-video prompt is doing a narrower job than a text-to-video one: the picture already fixes the subject, the framing and the lighting, so the prompt only needs to describe what changes. Keeping that distinction in mind is what separates a prompt that reads like a caption from one that actually directs motion.
Name the specific movement rather than the mood: "the model tilts her head and blinks" gives the reconstruction something concrete to animate, while "make it come alive" gives it nothing to work from and it defaults to whatever motion it has seen most often. Keep the action to one clear thing per generation, since a prompt asking for five simultaneous movements tends to blur all of them into a smaller, muddier version of each. A negative prompt is worth using whenever a first attempt comes back warped: naming what you do not want, like extra fingers or a shaking frame, steers the next attempt away from it more reliably than rewording the positive prompt again.
Pro tip: when a generation looks almost right but the motion is too aggressive, rerun the exact same prompt at the shorter duration option instead of rewriting it. A shorter clip gives the model less time to drift from the source photo, and it is often the fastest fix.
An honest description of this tool includes where it stops working. For turning a single photo into video, five limits show up often enough to plan around:
None of this rules the tool out. It rules out treating the first result as final without watching the whole clip.
Family photo restoration is the use case people describe most personally: a single still of a relative becomes a few seconds of them blinking or smiling, and for many people that is the first time they have seen the person move in years. It pairs naturally with old photo restoration work, since a cleaned-up scan animates more convincingly than a scratched original.
Real estate and product listings use camera-motion animation to turn a single hero photo into a short establishing clip without booking a videographer, a workflow that sits next to the fuller walkthroughs in real estate video tours. Social media creators use subject motion to turn a static portrait or flat-lay into something that stops a scroll, often as the opening seconds of a piece built with the techniques in AI product demo videos. Anyone comparing tools before committing to one can see how the model range here stacks up against a single-model service in the Picasso IA vs Kling AI breakdown.
Most photos work, but results vary with what the photo shows. A clear, well-lit photo with the subject fully in frame and not too small gives the model the most to work with, and portraits, products and landscapes all animate reliably. Photos that are already blurry, heavily cropped, or shot at an unusual angle tend to produce motion that looks slightly off, since the model has to guess at details the source image never captured clearly in the first place.
There is no single best choice, which is the point of having several. Kling v2.6 and Hailuo 02 handle natural subject motion like turning heads and gestures well and add synchronized audio. Wan 2.7 I2V supports locking both a first and last frame for a controlled transition. Ray 3.2 leans cinematic, with HDR output and six aspect ratios for wider shots. Running the same photo through two of them and comparing takes only a couple of minutes.
Most image-to-video models on Picasso IA generate clips of 5 or 10 seconds in one run, and some cap the longer option to a lower resolution. That is enough for a social clip, a product teaser or a single animated moment. For anything longer, the practical approach is generating several short clips and joining them with an editing tool, rather than expecting one generation to cover a full scene.
You get meaningful control, not frame-by-frame control. Your prompt sets the direction of the motion, and several models let you pin a start frame, an end frame, or both, so the clip opens and closes on images you chose. What you cannot do is hand-key an exact path the way you would in animation software; the model interprets your instruction and fills in the motion itself.
Some models will. Kling v2.6, for example, generates synchronized audio by default alongside the video, matched to what is happening on screen, and Wan 2.7 I2V can sync motion to an audio file you upload yourself. Others return silent video only. If sound matters for your clip, check the specific model's options before generating, since this varies model to model rather than being a universal feature.
It can, and doing a cleanup pass first usually helps. A photo with scratches, faded color or heavy noise gives the model a rougher starting frame, so the motion it adds can inherit those flaws. Running the photo through old photo restoration or a colorization pass before animating it typically produces a smoother, more natural-looking result than animating the damaged original directly.
Yes, some models support looping directly as a setting, producing a clip that cycles without a visible cut, which is useful for a website background or an ambient display. On models without a built-in loop option, setting the first and last frame to the same image gets you close to the same effect, since the motion returns to where it started.
It depends on the model, but most cover the platforms you would actually use. Common options span 480p through 1080p, and aspect ratios typically include 16:9 for widescreen, 9:16 for vertical social formats, and 1:1 for square posts. Setting the aspect ratio before generating, rather than cropping the finished clip, keeps the composition the model actually intended.
Each generation is paid in credits, and the exact cost depends on which model, resolution and duration you choose, since a longer or higher-resolution clip takes more compute. New accounts get free credits to try this before spending anything, and current plans live on the pricing page, the only place worth trusting for a real number.
Check two things first. That you have the rights to the source photo, since animating an image does not change who owns what it depicts. And the license terms attached to your plan, which govern how outputs may be used. As with any AI-generated content, give a commercial clip a human review before it ships, the way you would review a written or designed asset before publishing it.
A photo that has been sitting still for years is one prompt away from moving again. Open the Toolkit, upload one, and see what a few seconds of motion adds to it.
