One camera move
Two moves at once is the single most common cause of an incoherent clip: the model attempts both and commits to neither. If you want a compound move, generate two clips and cut between them.
Describe the motion, not the picture. Pick one camera move, one thing the subject does and one detail of the world, and this builds the prompt in the order a video model reads best.
FREE · NOTHING IS GENERATED HERE · 9 CAMERA MOVES THAT BEHAVE
Everything below composes text. Nothing is generated and nothing is sent anywhere.
THE METHOD
A motion prompt rarely fails for want of a better adjective. It fails in one of three specific ways, and this tool names each of them back to you rather than quietly rewriting your words.
Two moves at once is the single most common cause of an incoherent clip: the model attempts both and commits to neither. If you want a compound move, generate two clips and cut between them.
A clip that has to finish something inside its duration either rushes it or is cut off. "Candle flickering" behaves; "candle goes out" is a gamble against the clock.
The model can already see your start frame. Every word spent on what is in it is a word not spent on what should move — which is the only thing it needs from you.
PAST THE PAGE
Starting from an image fixes the composition, the colour and the subject before generation begins. That is why image-to-video usually takes fewer attempts than describing a whole scene from nothing — and why the prompt only has to carry the motion.
This page is free and needs no account. Generating video is a billed mode; the pricing page lists what a clip costs.
See how image-to-video worksQUESTIONS
Yes, and there is no account. It composes text in your browser — nothing is generated and no request is made, so it costs us nothing to serve. Generating the clip from the prompt afterwards is separate and billed.
Because a model given two will attempt both and commit to neither, and the result reads as a camera losing its balance. It is the most common cause of an incoherent clip. For a compound move, generate two clips and cut between them — which is also how it would be shot in reality.
Clips run 4, 6, 8 or 10 seconds and nothing in between. A request for 5 snaps down to 4, which is the usual reason an ending gets cut off. Six is the default and suits most single camera moves.
There is no standalone negative-prompt field in the Video Studio, so phrase things positively instead. "Static camera" works better than only saying "no camera movement", and both together work better still.
No — frame modes take frames only, and a separate style or character reference cannot ride along in the same request. Build the style into the source image first, which is cheaper anyway: fixing a look in a still costs an image rather than a video.
No prompt changes it, because it was never about wording. A photograph with an identifiable real person in it is held for a consent confirmation rather than refused — you confirm you have their permission and it renders. Generated portraits, illustrated characters, products, animals and places need no confirmation at all.
Longer clips, higher resolution, sound, songs and a timeline editor — in one account.
100 free starter credits in many countries · No card · Unused starter credits expire after 72 hours
No subscription: credit packs from €3.99, paid once. The credits you buy never expire.
Start creating freeWhat the studio expects from a start frame, and what it refuses.
No still to start from? Pull one out of footage, free and in your browser.
The general-purpose version, for prompts that are not about motion.
What generating from the prompt actually costs, in credits. No subscription.
Take it to the Video Studio, render it at 360p first, and change one thing at a time. That loop is the difference between a clip you post and a clip you delete.