AI Video Prompts: 8 Templates, Camera Moves & Fixes
Describe one readable shot before adding style. These eight editable AI video prompts cover settings, detail, character action, illustrated references, camera comparisons and prompt revision. Treat them as starting points, then inspect the output from the model you choose.
Scene templates open a video prompt draft. The revision assistant opens a conversation draft. No images are attached and no generation runs until you review the inputs and submit.
Make your first version
Prepare your references
- One idea that fits in a single shot, with a principal subject and action. Decide whether the task starts from text or an existing image.
- Your own or authorized image for an image-to-video template. This page does not attach a sample reference.
- A target aspect ratio and a limit on attempts. Check supported duration, inputs and pricing after choosing a model.
Steps in AI Fantas
- Choose a template that matches the task and replace every placeholder. Use a text-capable video model for text-only prompts, or upload your own reference for the image-to-video example.
- Put the subject, setting and action first. Add framing, one camera instruction and lighting, then remove competing actions or viewpoints.
- Open the draft, check the model, inputs, duration, aspect ratio and current credit quote, then submit it yourself. The last template enters conversation mode for revision rather than video generation.
- Watch the complete result, name the clearest problem and change one variable for the next attempt. Keep the original prompt, revision and output.
Open Studio Chat · Open Canvas · Read the help center
Adapt the prompt to your scene
These prompts are starting points rather than generated examples or guaranteed results. Upload your own reference for image-to-video. Text-only examples attach no image. Check the current model inputs, duration, aspect ratio and quote.
1. A reusable single-shot template
Try this prompt →2. Establishing shot: a coastal lighthouse
Try this prompt →3. Detail shot: a paper star with a locked camera
Try this prompt →4. Image to video: a breeze in your illustration
Try this prompt →5. Character action: a pause after the rain
Try this prompt →6. Camera comparison: the same star with a slow push-in
Try this prompt →7. An ending shot with room for the edit
Try this prompt →8. Prompt revision assistant (conversation draft)
Try this prompt →Check the result before using it
- Inspect subject identity, count, material and action throughout the clip, including invented objects and unwanted cuts.
- Separate subject motion from camera motion. A character turning is different from a camera orbit, and a subject approaching is different from a push-in.
- Pause at the beginning, middle and end. A prompt or reference cannot guarantee frame-by-frame continuity.
- Review actual generation settings, credit records and delivery needs. These templates promise no fixed price, runtime or visual result.
The six parts of an AI video prompt
A useful shot description usually names the subject, setting, action, framing, camera movement and lighting. The order can vary, but a clear event is easier to revise than a list of impressive adjectives. Words such as cinematic or realistic do not replace a readable action.
- Subject: one paper star, one lighthouse or one adult traveler, with a clear count and necessary appearance details.
- Setting and action: one principal location and one change. Separate a sequence that needs several locations into individual shots.
- Framing and camera: close-up, locked camera or slow push-in describe an intention, not exact mechanical control.
- Light and atmosphere: specify a direction and quality of light without combining contradictory conditions.
Write camera movement separately from subject movement
A locked camera makes an action easier to inspect. A push-in gradually moves closer, while a lateral move travels along one direction. Avoid asking for a fast orbit, zoom, pull-back and several character actions in the first attempt.
- Locked camera: keep the viewpoint still while one specified subject or environmental element moves.
- Push-in: say that the camera moves forward and clarify whether the subject remains still.
- Lateral move: name the direction and whether the subject should stay completely visible.
- Templates 3 and 6 use a similar paper-star subject and lighting with different camera instructions. They are an editable comparison setup, not a completed experiment proving a model response.
Text-to-video and image-to-video need different emphasis
A text-only prompt must establish the subject and environment. An image already supplies a composition, so focus on what moves and what should remain recognizable. Avoid describing a different character, pose or layout that contradicts the input image.
- Upload your own illustration before using the image-to-video template, and choose a model that accepts that kind of reference.
- Keep character, composition and style reference roles distinct. First-frame, last-frame and multiple-image support are different capabilities.
- A model invents information when a shot reveals an unseen back view or space. The prompt is not evidence of the real structure behind an image.
Fix the observed problem rather than adding more words
Name the most visible deviation and change the instruction that relates to it. If every attempt changes the model, subject, lighting and movement, it becomes difficult to learn what helped.
- Uncontrolled camera: remove competing movements and test the subject with a locked viewpoint.
- Deformation: reduce action, occlusion and turning; inspect the clarity of the reference.
- Too many actions or unwanted cuts: ask for one principal action in one continuous shot. Plan separate shots for a sequence.
- A crowded picture: remove competing subjects, props and background events.
- Unwanted text: add a brief constraint. Exact lettering is usually easier to typeset afterward than to reconstruct through generation.
Handle negative prompts and output settings correctly
Some models expose a separate negative-prompt field and others do not. Use a dedicated field only when the current interface supports it; otherwise keep essential exclusions brief inside the shot description. Set duration, aspect ratio and resolution through the actual model controls. Writing high resolution does not automatically select a higher output tier.
Should you write in English or Chinese?
Both versions of this guide provide copyable prompts. Start with the language in which you can describe and inspect the shot accurately. Do not assume a translation must improve the result. Clarify an unfamiliar camera term with a short explanation, then compare actual model behavior without silently adding subjects, movements or styles during translation.
Build a prompt log you can reuse
For accepted shots, retain the input image, prompt version, model, settings, attempts and actual credit record, plus any manual changes. When making a complete short film, return to the story and shot list before arranging editing and sound. A generated clip alone does not establish that the story, rhythm or continuity works.
If the first result needs work
Identify whether the issue comes from the description, reference, selected model or an overcomplicated shot. Use the revision assistant to address one problem. If the task only needs a pan or zoom across a still photograph, an editing tool may be sufficient without regenerating the scene.