How to Create AI Video: Text Prompts, Frames and Camera Movement
APIHub · Content reviewed:
Open the video category in APIHub Agents, sign in and choose an available video model. Describe the subject, action, scene and camera movement, then set duration, resolution and ratio. Add supported frames or reference images. Jobs finish asynchronously and settle on actual video tokens using the same balance as API calls.
Create a video in the browser without API integration
Open Agents video tools, sign in with your APIHub account and select an available model. No endpoint setup, API key entry or polling code is needed. Start with one short shot for an article opener, an event concept or a product scene.
Describe one clear shot
Include the subject, action, scene, camera movement, lighting, style and exclusions. Start with one main action and one camera move:
An unbranded white ceramic cup on a wooden desk near a window. A little steam rises in soft morning side light. The camera slowly pushes forward, keeping the cup central. Realistic style, simple background, no people, text, logos or sudden cuts. No dialogue needed.
Use the page controls for duration, resolution and ratio rather than relying on vague instructions in the prompt. Supported values depend on the selected model and current interface.
Choose text, frames or reference images
| Input | Useful for | What to describe |
|---|---|---|
| Text | Exploring a shot from scratch | Subject, action, scene and camera |
| Start or start/end frames | A given opening or transition between images | What to preserve and how the scene changes |
| Reference images | Subject or visual guidance | Consistent features and allowed changes |
Check which modes the model and interface support. For start/end frames, choose a plausible transition. A reference is guidance, not a guarantee of exact detail in every frame. Example instruction for a start frame: “Keep the cup shape, color and desk layout. Let steam rise slowly. Use a fixed camera; add no objects or readable text.”
Review the controls and budget hold
Choose a short duration first, an appropriate portrait or landscape ratio, resolution and the audio option you need. Do not assume generated speech will reproduce exact wording. Duration affects usage; resolution and video reference input affect applicable rates or usage.
Check the required hold before submitting. A positive balance can still be insufficient for a job’s budget, or other running tasks may already reserve available funds.
A hold is not the final charge
A hold reserves budget while the background job runs. Successful Seedance jobs settle using actual video tokens, the resolution/video-input price and applicable account/group rates:
Cost = actual video tokens / 1,000,000 × applicable unit price × account rate × effective group rate.
This is not a fixed per-second fee. An image reference is not video reference input. Check model prices and your applicable rates. Web video and API calls share one account balance.
Failed, expired or upstream-confirmed cancelled jobs release holds under the platform settlement rules. A timeout or a cancellation click does not prove the upstream job stopped. If submission is uncertain, check the original task or contact support rather than creating a duplicate paid job.
Wait for the asynchronous task and save the result
Tasks may be queued, running, successful or failed. Closing the browser does not automatically cancel a job. Keep its task ID and check history rather than repeatedly resubmitting.
View and download successful results promptly. Current result links typically last 24 hours and have download limits; the tool is not permanent video storage. If playback requires a supported codec, download and use a compatible player.
You can first draft a shot with the writing guide, develop a visual reference with the image guide, then test one short clip. Change one aspect at a time to judge the result and control unnecessary usage.
Ready to try it? Open Agents video tools, check model availability and prices, then start with a small job.