Make a still photograph move, or remake it entirely

What actually happens when a photograph is animated

A still photograph holds one instant. Everything a clip needs that the instant did not contain has to be produced from scratch: the frames on either side of it, the parts of the body the pose was hiding, the way fabric falls when an arm moves, the background that was behind the person before they stepped in front of it. The model is not extending the photograph. It is building a short sequence that is consistent with it.

That is why the picture matters more than the preset. A photograph that already contains most of what the action needs leaves very little to invent, and the result looks like the person in the picture doing the thing. A photograph that leaves out the legs a dance needs, or hides an arm behind a body, hands the model a much larger job, and every part of the job it does from scratch is a place the likeness can slip.

Nothing in this is a defect of one app. It is what image to video generation is, across every tool that offers it, and knowing it up front is the difference between a result you keep and a result you delete.

A clip, or the same picture remade

Half the catalogue sets the photograph moving and hands back a short clip of roughly 5 seconds. The other half remakes the photograph as a new still picture: a seventies film portrait, a chibi figurine, a noir frame, a poster. Both start from one image you already own, and they fail in different ways, so nothing on this site promises a video for something that returns a photograph.

A photo style is not a filter. A filter lays an effect over the pixels that are there; a style rebuilds the frame in that style, which means it can change what your jacket is made of, how the light falls and what is behind you. That is why a heavily stylised action drifts further from a likeness than a restrained one, and why the advice for choosing a picture is not the same on both halves of the library.

Every action page on this site reads which kind it is from the app's own catalogue rather than assuming, and says so on the page.

Generated, not recovered

This matters most with old photographs, which is what most people reach for first. A model that animates a photograph of your grandmother has never seen your grandmother move. It produces a plausible version of a person with her face doing the thing you asked for. It is not a recovery of a moment, and it is not evidence of anything.

People find that clip moving anyway, and there is nothing wrong with that. It is worth being clear about what it is, because the difference between an illustration and a record is exactly the difference between a keepsake and a false memory. We would rather say it plainly on the homepage than let the word "animate" carry an implication it has not earned.

If the photograph is an old print, repairing it first is the genuinely useful step, and it belongs before the animation rather than after it. A cleaner picture leaves the model less to guess at.

Why two runs are never the same

The result is generated rather than filmed, so the same photograph and the same action produce a different clip the second time. That is not a fault to be reported. It is the most useful tool you have: if the first attempt puts a hand somewhere strange or the movement reads slightly off, running it again often lands better without changing anything.

What a rerun will not fix is a structural mismatch. A crop that never included the legs an action needed will not grow them on the second attempt, and a face turned well past three quarters away will disappoint fairly consistently. The honest fix there is a different photograph, or a different action, which is what every page in this library is written to help with.