AI Video Creation - AnalyticDB - Alibaba Cloud Documentation Center

Feature Overview

The Video module converts screenplays or ideas into finished videos and serves as the finished video production engine of Creative Studio.

Feature	Use cases	Input	Output
Storyboard	Has a screenplay, wants systematic production from storyboard	Screenplay	Episodic finished videos
Infinite canvas	Free-form creation, node-based visual orchestration needed at any stage	Any assets, storyboard shots, prompts	Workflow + finished videos
Video replication	Has source video, needs derivative creation	Source video + reference materials + prompts	Rewritten short video
Quick creation	Creative prompts only	Prompts + character cards	Multi-scene assembled finished video
Video editing	Style/viewpoint transformation of existing video	Source video + prompts	Transformed video

Storyboard

Click Create from Screenplay on the Video homepage to enter.

Resource library

Automatic extraction: After creation from a screenplay, the system automatically extracts main characters, scenes, and assets.
Generate reference images: Check character and scene descriptions in the resource library and generate reference images, supporting individual or batch generation.

Storyboard generation

After clicking AI Generate Storyboard, the system automatically:

Splits shots by scene.
Automatically generates camera language, including shot types, camera movements, and shooting angles.
Binds character and scene resources to generate storyboard content.

Storyboard editing

Each storyboard card supports the following editing operations:

Modify scene descriptions, camera language, dialogue, and actions.
Add, delete, or rearrange storyboard shots.
Bind character and scene resources (@character name, @scene name references).
Set reference images and start/end frames.
Add annotations and status labels.

Video generation

Click Generate Video on a storyboard card.

Select a video model:

Model	Generation modes	Reference image	First frame	Start/end frames	Audio
Wan 2.7	Text-to-video, reference video, first-frame video, start/end frames	Supported	Supported	Supported	Supported
HappyHorse	Text-to-video, reference video, first-frame video, video editing	Supported	Supported	Not supported	Not supported

Supports reference start/end frames and multiple reference images.
Supports batch generation of multiple storyboard shots.

Post-editing

Episode assembly: Assemble all storyboard videos within an episode in chronological order. You can insert intros, outros, and transitions.
Voiceover and music: Select audio generated in the canvas or upload custom audio.
Subtitle synthesis: Generate subtitles for videos with one click.
Export format: Supports landscape 16:9 and portrait 9:16, with resolution of 720P or 1080P.

Infinite canvas

The infinite canvas provides a node-based visual workflow. You can freely place text, image, video, audio, and other nodes on the canvas, establish dependencies between nodes through connections, and the system automatically orchestrates and executes generation tasks in topological order. It is suitable for scenarios that require fine-tuning or free-form creation.

Canvas functions

Visualize storyboard shots and generation processes as nodes and connections.
Support drag-and-drop, combination, and reuse with WYSIWYG.
Bidirectional sync with the storyboard. Individual storyboard shots can be opened in the canvas for fine editing.

Node types

Video node

Instructions:

Click a video node to expand the prompt panel.
Write scene descriptions in the rich text editor, which supports the following features:
- Type @ to trigger the reference menu and reference connected character or scene nodes.
- Organize scene descriptions by time segments (such as "0-5s"), with each segment independently configurable for shot type, camera movement, and camera angle.
- Click the time label to edit the camera language parameters for each segment.
Select the model series, generation mode, aspect ratio, and resolution in the bottom toolbar.
Click the submit button (or press Cmd+Enter) to start generation.
After generation is complete, the video plays automatically within the node, with play/pause and progress scrubbing support.

Hover toolbar operations:

Action	Description
Extract last frame	Captures the last frame of the video as an image for continuous start/end frame generation.
Edit video	Opens the video editor for AI editing of existing videos using prompts.
Download	Downloads the video file locally.
Full-screen preview	Previews the video in a full-screen popup.

Image node

Instructions:

Click an image node to expand the prompt panel.
Enter an image description. Use @ to reference connected character/scene nodes as references.
Select the model, aspect ratio, resolution, and generation count.
Click Generate. After generation is complete, the image is displayed within the node.

Hover toolbar operations:

Action	Description
Crop	Freely crop the image or crop to preset ratios.
Inpaint	Use masks to mark areas in the image that need modification, and combine with prompts for AI inpainting.
Lighting	Adjust the lighting direction, brightness, and color temperature of the image through interactive 3D light orb controls.
Viewpoint	Generate new views from different camera angles based on a reference image.
Grid split	Splits the image into 2x2 / 3x3 / 4x4 independent nodes.
Outpaint	Extends content beyond the image boundaries to enlarge the image.
Remove background	AI background removal.
Annotate	Freely draw annotations and add text notes on the image.
Download	Downloads the image locally.
Full-screen preview	Views the image in full screen.

Audio node

Generation modes:

Mode	Description
Text-to-speech	Select a preset voice, enter text, and generate speech.
Voice design	Generate speech by describing ideal voice characteristics in natural language.
Voice cloning	Upload reference audio to clone the voice and generate speech.

Preset voices: The system includes 48 built-in voices covering different genders, ages, and personality traits. Each voice provides a name and personality description for selection.

Instructions:

Click an audio node to expand the prompt panel.
Select a TTS mode (text-to-speech / voice design / voice cloning).
In text-to-speech mode, select the target voice from the voice list.
Enter the text content to synthesize.
Click Generate. After generation is complete, the built-in audio player in the node loads automatically.

Text node

Text nodes are used for free-form text creation.

Instructions:

Double-click the node to enter edit mode, where you can directly enter or modify text content.
Click the node to expand the prompt panel, describe the text content to generate, and the AI generates text based on the description.
You can select different LLM models (Qwen3.7-Max, Qwen3.7-Plus, DeepSeek-V4-Pro, etc.).
The output of text nodes can be connected to image or video nodes to provide text context for downstream generation.

Resource library

The resource library is the user resource management center, supporting unified management of characters, scenes, and assets.

Open the resource library

Click the "Resources" button in the left toolbar to open the resource panel.

Browse and filter

The resource panel provides filtering in three dimensions:

Type tags: All / Characters / Scenes / Assets / Audio / Style / Others
Project scope: All projects / Current project / Custom selection
Keyword search: Supports fuzzy search by name
Favorites filter: Display only favorited resources

Create resources

Click the "+" button in the resource panel.
Select the resource category (Characters/Scenes/Assets/Audio/Style/Others) in the folder selector.
Select files to upload.
After resource creation is complete, you can manage them in the panel.

Use resources

Drag resources from the resource panel onto the canvas. The system automatically creates nodes of the corresponding type.

Management operations

Upload: Supports dragging local files directly into the canvas for automatic upload and node creation.
Categorize: Automatically categorized by character, scene, style, reference image, video, and audio.
Favorite: Pin frequently used assets to the top.
Search: Search by name, tag, source, and reference count.

Workflow templates

Workflow templates are reusable node and connection configurations that allow you to quickly set up commonly used creative workflows.

Use templates

Click the "Templates" button in the left toolbar to open the workflow panel.
The panel has two tabs:
- Public templates: System preset templates and shared templates.
- My templates: Templates you created.
You can search for templates by keyword.
Click a template card to load it onto the canvas with one click.

Create custom templates

Set up the workflow on the canvas (nodes and connections).
Select multiple nodes and click "Group" in the multi-select toolbar.
Click "Publish Template" in the group node toolbar.
In the publish dialog box, enter the name, tags, save location, and description.
Click "Publish".

Typical creative workflows

Scenario 1: Text-to-image-to-video full pipeline

Add a text node and enter or AI-generate screenplay/scene descriptions.
Connect the output of the text node to an image node, using the text as context for image generation.
In the image node, set the model, aspect ratio, and style to generate static images.
Use the editor to refine images (crop, inpaint, adjust lighting, etc.).
Connect the image node to a video node as a reference/first frame for the video.
In the video node, set the camera language (shot type, camera movement, camera angle) and duration to generate the video.

Scenario 2: Character-consistent video series

Create a character node and upload or generate a character reference image.
Use the viewpoint feature to select different angles through 3D camera controls and generate multi-viewpoint reference images.
Create multiple video nodes, each describing a different plot shot.
Connect the character node to all video nodes.
The system automatically passes the character reference image when generating each video, ensuring consistent character appearance.
Select all nodes and group them, then click "Execute Group" for batch generation.

Scenario 3: Quick start with workflow templates

Open the canvas and select "Browse Templates" in the welcome dialog box.
Browse preset templates in the workflow panel and select a suitable one.
The template loads automatically onto the canvas, including preconfigured nodes and connections.
Modify the prompts, character, and scene information in the nodes.
Execute the workflow to generate all content with one click.

Video replication

Video replication generates new versions based on existing videos and prompts, suitable for derivative creation of style, plot, or characters from existing materials.

Upload the source video.
Add reference materials: Upload reference images, reference videos, or character cards to constrain the art style and character appearance of the new video.
Enter creative prompts: Describe the desired rewriting direction, such as "Convert the original video to an ancient costume style while keeping the camera work and pacing."
Configure generation parameters.
Click Start Generation.
Preview the results after the task is complete. You can download, add to the asset library, or continue editing.

Quick creation

Quick creation is suitable for scenarios of automatic long video generation, such as live-action voice-over advertisements, live instructor courses, animated short films, or free-form creative scenarios.

Enter a creative description, such as "A programmer couple meets, argues, and reconciles against a city night skyline."
Select the target duration: 15 seconds, 30 seconds, 45 seconds, 60 seconds, or 90 seconds.
Select the aspect ratio: 9:16, 3:4, 16:9, 4:3, or 1:1.
Select the generation mode (parallel / sequential) and speech rate.
Click Start Generation. The system will:
- Automatically split into multiple scenes based on the target duration.
- Generate prompts and visuals for each scene.
- Batch-generate video clips.
- Automatically assemble into a complete short film.

Video editing

Video editing is used for re-creating existing videos with style transformation, viewpoint changes, domain randomization, and other modifications, as well as for asset derivative creation.

Upload or select a source video.
Add up to 5 reference images (optional).
Describe the target effect, such as "Change the lighting to warm sunset tones while keeping the camera position."
Click the configuration button below to set:
- Resolution: 720P or 1080P.
- Audio processing: Automatic dubbing or keep original audio.
Optionally use one-click prompt optimization, where the system automatically rewrites prompts for better precision.
Click Start Generation and wait for the task to complete.