AI Video Generation
Use SeaArt AI video generator to turn text, reference images, videos, or audios into short videos.
Overview
You can open the SeaArt AI video generator in either of these ways:
On the Home page, click Video Generation.
Click Create in the left sidebar and select Make Videos.
Click Create in the top navigation bar and select Make Videos.

Use the "Video Model" tab to generate a video from text, images, or more reference media. The top of the workspace also has "History" and "Inspiration" tabs. Use History to find the videos you have generated, and Inspiration to browse examples from the SeaArt community.

Video Generation Modes

Text to Video
Enter a prompt that describes the scene, subject, motion, camera angle, and style. SeaArt generates a video from your description.
Image to Video
Upload an image as the starting or ending point, then add a prompt to describe how the scene should move.
Reference to Video
Reference to Video lets you provide images, videos, and audio as references for the model to follow. Each asset can control a different part of the result, such as character identity, visual consistency, motion, camera movement, or sound.
@Image - use images to keep characters, props, costumes, scenes, or visual style consistent. Images can also define the starting and ending frames when needed.
@Video - use a reference video for action or camera movement.
@Audio - add an audio clip to guide the sound or rhythm.
You can combine multiple reference types in one prompt. For example, use @Image 1 to keep the character's appearance consistent, @Video 1 to apply a camera movement, and @Audio 1 to set the rhythm of the scene.
Edit Video
Upload a 4- to 30-second video, then use text or image inputs to edit the original footage, switch perspectives, and change the visual style.
Video Output Parameters
Model
Select the AI video model to use. Each model supports different resolutions, durations, and input modes.
Resolution
Choose 480p, 720p, 1080p, 2K, or 4K. This setting is labeled Generation Mode in the interface. Available options depend on the model.
Aspect ratio
Choose 16:9, 9:16, 4:3, 3:4, 1:1, or 21:9. Available ratios depend on the selected model.
Duration
Set the video length. Available durations depend on the model.
Number of videos
Set the number of videos to generate per run.
Prompt Magic
Controls how SeaArt enhances your prompt. Choose Auto, On, or Off.
Private creation
When enabled, your generated video is kept private.
Some models may offer Limited-Time Free generation during promotional periods. Availability is shown in the model panel.
Generate a Video

Select a video model from the model panel.
Choose an input mode: Text to Video, Image to Video, Reference to Video, or Edit Video.
Enter a prompt or upload the required images or videos.
Adjust the output settings. Available options depend on the selected model.
Click Generate.
View your video in History tab.
Common Advanced AI Video Models
The models available on the Video page are updated regularly. The following models cover a range of workflows, from short social clips to multi-shot cinematic sequences.
A multimodal video model from ByteDance's Seedance family. It supports text, image, video, and audio references, generates clips up to 30 seconds with synchronized audio, and outputs up to 4K. Use it for ads, short films, and multi-shot sequences that need consistent characters and detailed camera direction.
A cinematic video model from ByteDance's Seedance family. It supports text-to-video and image-to-video, follows structured director-style prompts, and can generate clips up to 60 seconds with native audio. It does not currently support real-person face references.
A general multimodal model that can generate, reference, and edit audiovisual content in one workflow. It accepts up to 9 images, 3 video clips, and 3 audio clips, produces 5- to 15-second videos at 768p or 2K, and supports character replacement, background changes, and voice cloning. It is best suited to dialogue-driven scenes, short dramas, and product showcases.
Alibaba Cloud's multimodal video model. It accepts text, images, video, audio, documents, and public webpages, supports first- and last-frame control, and generates videos up to 30 seconds at up to 1080p with audio on by default. It works well for product stories, presentation openers, and before/after transformations.
A cinematic text-to-video and image-to-video model from Kuaishou. It supports storyboard-style prompts with up to six shots, image or video references for character and style consistency, native audio, and camera-direction prompts. Clips run from 3 to 15 seconds at up to 1080p.
Google's multimodal video model for creating and editing through natural conversation. It accepts text, images, audio, and video references, and lets you refine a scene by changing the background, objects, lighting, camera movement, or style without starting over. It also supports audio and style transfer.
Txt2Vid
Text to VideoImg2Vid
Image to VideoLast updated