Organize recurring characters, props and locations into reusable assets so shots share a consistent set of references.
Create assets
Open Assets, choose Characters, Props or Locations, then add an asset with +. Give it a distinct name and description, such as “Lin,” “Brass pocket watch” or “Old newsroom.”
You can also use the AI assistant at the lower right to extract characters, props or locations from the current episode’s screenplay. Review names and descriptions. For duplicates, choose whether to skip, update or create a separate asset. Candidates become project assets only after you apply them. Extracting text does not also generate the images.
Understand the workspace
Select an asset and manage retained versions on the left. View its image, write prompts and use tools in the center. Browse generation history on the right. Hover over icon tools to see their names.
Characters have basic-appearance and look areas. Props and locations have main-image and supporting-image areas. Check the active area before generating or selecting results.
Generate and adjust
- Select the asset and area, then refine its description or prompt.
- Add local references or project assets and check the model’s input support.
- Choose the model, quality, quantity and style, then generate.
- Open results in history, compare them and retain the versions you need.
Establish a character’s basic appearance before creating looks. Establish a prop or location’s main image before making supporting views. For local changes to a 2D image, use Modify and follow that editing flow.
The style control can override the generation style for this asset or restore the project style. It affects future generation; it does not re-render existing images or change the project style for everyone.
History, versions and final selection
History is chronological: older batches are at the top and the latest at the bottom. The list initially opens at the top; starting generation scrolls to the bottom. A 3D location’s in-progress card also appears at the bottom.
| Action | Result |
|---|---|
| Generate | The result first appears in history |
| Add to versions | Retain it in version management for review and comparison |
| Finalize | Confirm a retained version as the asset’s final choice |
| Remove from versions | Stop retaining it there without deleting its history file |
| Delete | Delete the version; read the confirmation first |
Viewing an image is not finalizing it. Unfinalize a final version before removing it where required by the UI.
2D and 3D locations
In the main-image area, choose 2D or 3D generation. Use 2D for atmosphere and composition references; use 3D for spatial inspection and camera planning. Their previews and tools differ. A 3D scene does not use the 2D local-image modification tool.
Character voices
Open Voice from the character tool panel in either its basic-appearance or look area. The voice belongs to the character, so changing a look does not change it.
- Open the character’s Voice dialog.
- Choose Project voices, Official presets, AI design, or Record/upload.
- Listen to or prepare a sample, confirm its name and bind it to the character.
- In Video, check the voices enabled for this generation.
| Source | How it works |
|---|---|
| Project voices | Reuse a voice already saved in the project; multiple characters can select it |
| Official presets | Select a preset and prepare its reference sample; selection saves it in the project voice library |
| AI design | Describe the sound, provide sample text and choose a model to generate the voice and sample |
| Record/upload | Use 2–15 seconds of clean speech; record with the microphone or upload and trim a file |
A recorded or uploaded sample can be played and used as a video voice reference directly. Cloning is required only when you want it to speak newly entered text. Confirm that the speaker has consented before cloning. You can then choose a speech model and reading settings, and use a reading as the voice’s new reference sample.
AI design, cloning and text-to-speech are paid operations. Check the model information and usage records. Speech is charged proportionally to the actual character count, rather than rounding every short request up to a thousand characters.
Voices and bindings are shared within the project. You can change or clear a binding. To delete a project voice that is still in use, first change or clear the bindings of the characters using it. Changing a sample or binding, or deleting a voice, does not rewrite previously generated videos.
Video with character voices requires a supported mode and audio-reference-capable model. See Video.
Related: Shot lists and storyboards · Video · Glossary