AI video prompt generator: fill-in templates for Veo, Kling and Seedance

A generic prompt rewriter gives you one block of text for any model. This page gives you a manual generator instead: six slots to fill, then a template for Veo, Kling or Seedance that follows each maker's own guide, read on 2026-10-26.

How do I use this AI video prompt generator?

You fill six slots once, then copy the slots into the template for your model. No signup is needed, and nothing is sent anywhere.

Write one short line for each slot:

1. Subject. Who or what is on screen. Add two or three details such as clothing or color.
2. Action. One main thing that happens. Use a verb and a count, such as "pours milk, then lifts the cup".
3. Place and light. Where and when, and the light source.
4. Camera. One framing and one move, such as "medium close-up, slow push-in".
5. Sound. Effects, background noise and any spoken line. Say "no music" if you want none.
6. Avoid. Things you do not want, written as nouns.

Keep one camera move and one main action per clip, as in the prompt examples. Kling's guide says a clean script works better than a complicated structure, and Google says you do not need every element in every prompt. Kling 4.0 accepts prompts up to 8,000 tokens, but a larger limit is not advice to write a long prompt.

What does each model want in a prompt?

The three guides differ mainly in suggested slot order, how dialogue is written and what clip length allows. The table compares the guidance read on 2026-10-26.

Neither Kling's blog nor fal's guide gives an avoid-list rule, so the Kling and Seedance templates leave that line out.

Veo 3 prompt generator: the template

A Veo prompt works as plain sentences that follow Google's element list, with audio in its own sentences. Fill the brackets from your six slots.

[Camera framing and move]. [Subject with details] [action]. [Place and time], [light]. Sound effects: [sounds]. Ambient noise: [background]. [Speaker] says: [line].

Google's guide gives audio as three types: sound effects, ambient noise and dialogue. For an avoid list, add a separate line of nouns such as "wall, frame", not "no walls". The Google Cloud page lists veo-3.1-generate-001 with clip lengths of 4, 6 or 8 seconds and a retirement date of November 17, 2026 or later, so confirm which Veo version your tool runs. If audio is missing, see Veo 3 not generating audio.

Kling prompt generator: the template

A Kling prompt is one paragraph of plain scene text, with each spoken line tied to a named speaker. Kling's guide says a clean script works better than a complicated structure.

[Subject with details] [action], in [place and time] with [light]. [Camera framing and move]. [Speaker] says [delivery]: "[line]." Sound: [effects and ambience].

Kling's advice is to name the speaker, write the line and describe the delivery, and to break long sentences into shorter ones. Use delivery words such as whispering or hesitant only when they matter. For image input and end frames, see Kling image to video.

Seedance prompt generator: the template

A Seedance prompt starts with the subject and its motion, then adds optional layers. fal's guide calls subject and motion the two parts that do the heavy lifting.

[Subject with details] [motion]. [Place and time], [light], [look]. [Camera framing and move]. [Sounds], no music. [Speaker] says, "[line]."

Put spoken words in double quotes and keep them short, because fal says long monologues drift out of sync. fal suggests testing a style at 5 seconds and 720p before longer runs. If you add reference files, the rules are in how to use Seedance 2.0.

Image to video prompt generator: the template

An image to video prompt describes only what moves and how the camera behaves, because the uploaded picture already shows the subject. Higgsfield advises a sharp, front-facing, well-lit image and a short prompt for the action and the camera move.

[What moves, one verb]. [Secondary motion such as steam, hair or cloth]. [Camera move]. Sound: [effects].

Higgsfield also says to start with image to video before text to video, and to keep the first generation under 8 seconds. On Veo 3.1, Google's page says reference-image-to-video supports 8 seconds only, so a 4 or 6 second clip needs a different input mode.

One idea, built four ways

The idea is "a barista pours latte art at 7 a.m." The four prompts below are built from the templates. Word and character counts include everything shown.

Veo 3.1 (61 words, 333 characters):

Medium close-up, slow push-in. A barista in a green apron pours steamed milk into a latte and forms a leaf pattern. A small cafe at 7 a.m., warm window light, soft haze. Shallow depth of field. Sound effects: milk hissing, a cup set on a saucer. Ambient noise: low cafe chatter and an espresso machine. The barista says: Here you go.

Kling 3.0 (55 words, 287 characters):

A barista in a green apron pours steamed milk into a latte and draws a leaf pattern, in a small cafe at 7 a.m. with warm window light. Medium close-up, slow push-in. The barista says softly, with a small smile: "Here you go." Sound: milk hissing, a cup set on a saucer, low cafe chatter.

Seedance 2.0 (52 words, 278 characters):

A barista in a green apron pours steamed milk into a latte and draws a leaf pattern. Small cafe at 7 a.m., warm window light, shallow depth of field. Medium close-up, slow push-in. Milk hissing, a cup set on a saucer, low cafe chatter, no music. The barista says, "Here you go."

Image to video, from a photo of the barista (20 words, 121 characters):

The barista tilts the pitcher and the leaf pattern forms. Steam rises. Slow push-in. SFX: milk hissing, low cafe chatter.

The same six slots produced all four. What changed is the order, where the camera sits, how the line is written and how much the prompt repeats from the image. Moving an old Sora 2 prompt to another model? Refill the six slots with that model's template instead of pasting it over.

How do I check a prompt generator's output?

PrompTessor, one top result for this query, lists Veo 3.1, Runway Gen-4.5, Kling 3.0, Seedance 2.5, Wan 3.0 and MiniMax H3, and says its workflow is checked against provider guidance (page read 2026-10-26). Whichever tool you use, run its output through four checks:

• Does the dialogue format match the model (quotes, says:, or speaker plus delivery)?
• Is the prompt short enough for one action and one camera move?
• Is the clip length you asked for allowed by the model?
• Does it avoid "no" and "don't" in avoid lists on Veo?

Can I use real people or characters in a prompt?

Don't name a real person, use a real face, or copy a named film or cartoon character unless you hold the rights. Platforms set their own face and rights rules, so read the current terms of the tool you publish with before you upload a face or name a character.

What are the limits of these templates?

• Seedance rules come from fal's guide, a model host, because BytePlus's own Seedance 2.0 prompt guide loaded as an empty shell on 2026-10-26. Reference-file syntax is outside this page.
• Higgsfield's guide uses its own @character style tags, which are platform syntax and are not used in the templates above.
• Kling's guide does not give image-to-video rules, so the image template follows Higgsfield's advice.
• Model versions and retirement dates change. The Veo 3.1 page showed a retirement date of November 17, 2026 or later on 2026-10-26. To choose a model first, use the AI video model finder.