Video to prompt
Any clip in. The prompt out.
Upload a video and AI watches it, sound included, then writes the prompt to recreate it: subject, camera moves, lighting, pacing and audio. Ready for Seedance, Kling, Veo, Sora or Wan.
Free with an imageat account. Upload clips you have the right to use.
Red Coat in Tokyo Night
A young woman in a vibrant red coat walks confidently through a neon-lit Tokyo alleyway at night.
A young Japanese woman with long black hair, wearing a long red wool coat over a black dress, walks toward the camera through a narrow Tokyo side street at night. The ground is wet and reflective from recent rain. On the left, a street food stall with steaming pots and a cook in a black uniform is visible. Neon signs in blue, pink, and white glow intensely, casting colorful reflections on the pavement. The camera performs a slow tracking shot backwards, keeping the woman centered as she walks with a calm, neutral expression. Cinematic 35mm film style with deep shadows and high-contrast neon lighting. Ambient city sounds, distant chatter, and a low-tempo lo-fi hip-hop track.
We made these clips. It read them back.
Each clip was generated with Wan 3 from a prompt we wrote. Video to Prompt saw only the video and wrote its own. Flip between the two to see how much it recovers, then generate from either.
A close-up selfie-style shot of a woman with curly brown hair and a denim jacket over a white t-shirt, holding a ceramic cup of latte. She speaks enthusiastically to the camera, smiling and gesturing with her head. The setting is a bright, modern cafe with large windows, wooden furniture, and warm pendant lights. The background shows other patrons and a barista. The lighting is soft and natural, coming from the large window to the right, creating a warm, inviting atmosphere. The camera remains steady in a medium close-up, capturing her expressive facial movements and the steaming coffee cup. High-quality digital video with a shallow depth of field, keeping the focus sharp on the woman's face.
Medium close-up selfie-style shot; the camera is slightly shaky but generally stable, mimicking a handheld phone recording.
A woman's voice says, 'Okay, I finally found the best coffee in the city, and you have to try the oat milk flat white.' Background cafe ambience includes light chatter and the hum of an espresso machine.
Eight things it writes down
It watches the motion itself, not a few stills, so camera moves and cut timing come from what actually happens on screen.
Subject
Who or what is on screen: appearance, wardrobe, materials.
Setting
Location, time of day, weather and set dressing.
Action
What happens, beat by beat, in order.
Camera
Shot size, angle, lens feel and every move: push-in, orbit, crane.
Lighting
Key direction and quality, colour temperature, contrast.
Style
Phone footage, 35mm film, 3D, anime, and the colour grade.
Pacing
One continuous take or fast cuts, real time or slow motion.
Audio
Quoted dialogue, music, sound effects and ambience.
How to turn a video into a prompt
Upload a clip
Drop an MP4, MOV or WebM up to 60 seconds. Vertical, square and landscape all work.
Pick the model
Choose Seedance, Kling, Veo, Sora, Wan or any model. The prompt is written the way it reads best.
Copy or generate
Copy it as a paragraph, shot list or JSON, or open it in the video studio with the clip's ratio preset.
From reference
to your own video
Recreate a format
Found a hook, transition or product shot that works? Get its structure in words, then shoot or generate your own version with your subject.
Localise a winner
Keep the camera, pacing and sequence of a proven ad, and swap the setting, language, product or cast in the prompt before you generate.
Plan variants per model
Analyse once per target and run the same idea on Seedance, Kling and Veo side by side to see which model handles the shot best.
Learn to prompt video
See how a clip you like breaks down into camera language, lighting and pacing. It's the fastest way to learn the vocabulary models respond to.
One clip. Six dialects.
Video models read prompts differently. Pick the model before you analyse and the paragraph follows its conventions. The breakdown, shot list and JSON stay the same.
Seedance
Direct sentences, explicit camera moves, "Cut to:" between shots.
Kling
Subject → action → scene → camera → lighting, in cinematography terms.
Veo
Rich visuals followed by dialogue, sound effects and music.
Sora
Scene, then cinematography, then the action beat by beat.
Wan
Spoken lines in quotes and key sounds for its synced audio.
Any model
A balanced prompt that travels well between models.
Get the prompt
behind any clip
Upload, copy, generate. Then make it yours in the imageat video studio with every top video model in one place.