AI Video Generator

Two of the three modes are compared here: writing a scene from text, and casting it from reference pictures. Prices are for five seconds; the longest take comes from each model's own form.

Starting from a picture? Image to video has its own comparison, with end frames and one-tap effects. Compare image-to-video models

• UncensoredRuns with no content filter added on our side.
• LightPart of the range passes, part is still filtered.
• FilteredRequests beyond the model's own limits are refused outright.

Real prompts from text and from references, each with the clip it made. Notice the order they follow: subject, action, camera, light, sound.

• 

  Walking on a mirror-flat salt flat at blue hour

  Wan 3.0 PrimeNature & landscapes

  A mirror-flat flooded salt flat at blue hour, sky and ground identical so the horizon disappears. An adult woman walks slowly through two centimetres of water toward the camera in a translucent white slip dress, barefoot, a long sheer scarf dragging behind her across the surface; her perfect reflection walks upside down beneath her. She stops and spins once, sending rings across the mirror. Wide static frame that then pushes slowly in. Pale lavender and cyan gradient with one warm rim of last light. Surreal and hushed. No text, no logos.

  Copy promptWan 3.0 Prime model page
• 

  Crimson silk gown sinking through blue water

  Wan 3.0 Pro PrimeFantasy

  Underwater slow motion, natural light shafts spearing down from the surface. An adult woman in a flowing crimson silk gown sinks slowly through deep blue water, the fabric blooming and twisting around her, hair fanning out, a stream of bubbles trailing from her fingertips as she reaches up toward the bright surface. The camera drifts in a slow arc around her and tilts up. Deep blue to teal grade, god rays, weightless, cinematic, 50mm.

  Copy promptWan 3.0 Pro Prime model page
• 

  Camel coat hero turn in 4K

  Wan 3.0 ProProduct & ads

  Keep the coat from the reference exactly — cut, colour and stitching. The model turns once on a dark stage and the hem lifts and settles. Slow orbit, single hard key, no music.

  Copy promptWan 3.0 Pro model page
• 

  Sneaker turntable from three product photos

  Seedance 2.0 FastProduct & ads

  Treat @Image1, @Image2 and @Image3 as three angles of one sneaker. Rotate it smoothly on an invisible turntable through those three views in order, keeping the leather grain, gum sole and lacing identical throughout. Hard studio key stays fixed; the contact shadow swings with the rotation.

  Copy promptSeedance 2.0 Fast model page
• 

  Put one character into a given room

  Wan 3.0Architecture & interiors

  Image 1 is the character — keep her face, hair and the black slip dress exactly. Image 2 is the room. She crosses from the doorway to the window and leans against the frame, looking out. Slow lateral track, one practical lamp, night ambience.

  Copy promptWan 3.0 model page
• 

  Lone swimmer in an empty night pool, cut to a beat

  Seedance 2.5Architecture & interiors

  Follow the camera language of the reference clip — low, wide, drifting right — but stage it in an empty indoor pool at night. Cut the movement to the pulse in the reference track. One swimmer, underwater lights, tiles echoing.

  Copy promptSeedance 2.5 model page

1. 01

   Choose how to start

   From an image locks the look. From text is the quickest. From references keeps the same face, outfit or product across several shots.

2. 02

   Direct the shot

   One subject, two or three beats, one camera move, then the light and the sound. Every model here reads that order well.

3. 03

   Draft short, finish long

   Run five seconds at the lowest resolution to check the motion, then render the take you like at full length and size.

• Give the camera its own sentence: "Slow push-in." "Handheld, following her." "Locked-off wide." Models mix up camera and subject moves when they share one.
• Two or three beats per five seconds, joined with "then". One small action stretched across a whole clip plays in slow motion.
• Describe the sound as well. MiniMax H3, Seedance and Wan render audio in the same pass, so "rain on the glass, distant traffic" ends up in the file.
• Write what you want, not what to avoid. A "don't" line tends to freeze the motion along with the thing you were trying to stop.

• Packing a whole story into one clip. Five seconds holds two or three beats; ask for six and the model rushes some and drops the rest. Split the story across clips.
• Describing a face in words and hoping it comes back. Text-to-video recasts the person on every run. When a character has to return, cast them from reference pictures or start each clip from the same still.
• Forgetting the sound. Some models add audio to every clip and others have it switched on by default, so unasked-for music is common. Name what should be heard, and write "no music" when you want quiet.
• Rendering the final before the draft. A long take at full resolution is the most expensive run on this page, so check the motion at five seconds and the lowest resolution first.

We add no content filter of our own. Some models on this page soften or refuse some requests by themselves, and their badges say so.

• Nothing involving minors, or anyone who could be read as one.
• No sexual or intimate content of a real, identifiable person without their documented consent.
• Nothing that is illegal where you or your audience are.

A five-second clip here runs from $0.06 on P-Video 2 Pro at 480p to $3.41 on Seedance 2.5 at 1080p; longer clips scale by the second.

Everything on this page is one API call. MiniMax H3 takes the same parameters in code as in the form above, at the same price per run.

• Text, image or reference: when each Wan 3.0 mode winsHow to choose the mode by the one thing in the shot that must not change.
• Every video model's five-second price, side by sideThirty-four models at three resolutions, and when paying per clip beats a plan.
• MiniMax H3 prompt guideThe six-part prompt formula, weak and strong pairs, and fixes for frozen motion.
• MiniMax H3: our hands-on review
• Seedance 2.5: our hands-on review
• Wan 3.0: our hands-on review
• Video models ranked
• Text-to-video prompt library
• The canvas: plan the whole piece