Action and Camera Movement Prompts: Making the Camera Move
When to Use
You have already gotten your first shot working, but the result feels like "it moves, but it doesn't look good": the action range is too small, the camera is stiff, the atmosphere is flat. This page teaches you how to turn every shot in your storyboard into an executable AI video prompt, so that shot size, action, and camera movement are all generated according to your intent.
This page uses the Wan2.2 image-to-video workflow as an example; the writing approach also applies to text-to-video.
Before You Start
- You have completed the prerequisite page "First Shot: Closing the Loop with Image-to-Video": ComfyUI can find the template "Wan2.2 5B" and successfully produce a clip.
- You have a storyboard at hand, with each shot at least specifying "who, where, doing what".
- If you want to use the official prompt expansion, prepare a Dashscope API Key (apply at Alibaba Cloud Model Studio following the official README instructions), or deploy the Qwen series locally.
Follow Along: Write Action and Camera Movement in Three Steps
Step 1: Break Down the Official Example and Build the AI Video Prompt Skeleton
First, look at an example prompt from the official README that can serve as a model: a white cat with a surfboard on a summer beach. Taken apart sentence by sentence, it is organized in four segments:
- Shot size: close-up;
- Subject and action: the white cat and the surfboard;
- Background elements: details of the beach environment;
- Atmosphere: overall lighting and mood.
Rewrite your storyboard following this skeleton: first set the shot size, then write what the subject is doing, then add the background, and finally the atmosphere. One sentence per segment, with few piled-up adjectives.
Step 2: Apply the Camera Movement and Action Vocabulary
Whether the camera moves depends on whether the prompt contains an explicit camera movement instruction. The Chinese-English vocabulary below is a teaching template designed for this course, not an official vocabulary:
| Camera Movement | How to Write the Prompt |
|---|---|
| Static | static camera, locked shot |
| Push in | camera slowly pushes in |
| Pull back | camera pulls back |
| Truck | tracking shot, camera moves sideways |
| Follow | camera follows the subject |
| Orbit | camera orbits around the subject |
Write actions with a clear range and direction: instead of "she moves", write "she slowly turns her head to look out the window". It is recommended to write only one primary camera movement at a time to avoid conflicting instructions.
The official Wan2.2 Prompt Guide contains a more complete set of camera language writing conventions; that guide is an online document not snapshotted by this site, so the official terminology prevails.
Step 3: Enable Prompt Expansion
A short-sentence skeleton can be expanded into a full description using the official expansion mechanism:
- Cloud: Dashscope API, use
qwen-plusfor text-to-video andqwen-vl-maxfor image-to-video; - Local:
Qwen2.5-14B/7B/3B-Instruct(text-to-video) orQwen2.5-VL-7B/3B-Instruct(image-to-video); - Add
--prompt_extend_target_lang 'zh'to target the expansion at Chinese.
Note: these parameters come from the official repository's command-line usage; whether the ComfyUI workflow exposes equivalent switches, please verify in your nodes.@@I18N_KEEP_000009@
Checkpoints
Self-check each item before generating:
- The AI video prompt includes the four segments: shot size, subject action, background, and atmosphere;
- Only one explicit primary camera movement is written;
- Actions have a range and direction, without vague words like "moves" or "walks";
- When writing in Chinese, the expansion target language has been confirmed to avoid mixing Chinese and English.
Boundaries
- Wan2.2's "professional camera language" is the official marketing claim; actual results depend on your own generations.
- The camera movement vocabulary is a course template; the official Prompt Guide is not snapshotted by this site, and the terminology there follows the official version.
- A well-written prompt does not guarantee a usable clip on the first try: broken actions and camera drift can still happen; for troubleshooting, see "Shot Failure Diagnosis: Pinpoint the Problem in 60 Seconds".
Next Steps
- Want precise control over where the shot starts and ends: go to "First and Last Frames Control Where the Camera Goes".
- Inconsistent style across the whole film: revisit "Style Consistency: Making the Whole Film Feel Like One World".
- Need finer 14B text-to-video configuration: see the ComfyUI section's Wan2.2 14B T2V tutorial.