Storyboards and generation

The Storyboards workspace organizes an episode into producible shots. Each shot has a prompt, references, generation settings and video candidates. Establish a reliable single-shot setup before running a batch.

Open and inspect a shot

1. Complete script and asset settings, then select an episode and scene in Storyboards.
2. Select a shot from the list or thumbnail strip.
3. Read its action, dialogue, framing and duration. Confirm that it connects with the surrounding story.
4. Verify characters, locations, props and voices. A reference must point to an asset, not merely contain the same name as plain text.
5. Choose the model, ratio, resolution, duration and available candidate count.

Review shots in the storyboard workspace before individual or batch generation.

Review shots in the storyboard workspace before individual or batch generation.

Choose a generation approach

Multimodal references combine identity, location, audio and movement guidance. Give each reference a specific responsibility in the prompt.

First and last frames suit continuous camera movement, transformations, entrances, exits and controlled endpoints. For multiple cuts, use explicit time segments or keyframe planning instead of asking two fixed endpoints to carry a complex sequence.

Edit prompts and mentions

Use @ to select references, or click an asset name in the sidebar. Pasted text can automatically match asset names; still inspect ambiguous names, costume variants and voice assignments afterward.

Extract blocking descriptions from a reference image to improve left/right relationships and framing between shots. AI-assisted prompt conversations can help revise movement, camera direction or wording. Read the complete revised prompt before generating.

Test, then generate a batch

Choose a test shot that represents the production's difficulty—for example, two-person dialogue or moving blocking. Once identity, voice, timing and quality are acceptable, batch-generate reviewed shots. Check the shot count and parameters first.

For failures, read the error and keep its ID. Avoid repeated clicks while a task's status is uncertain. Correct the inputs or retry only after you know the outcome.

Select historical versions

A shot may have several candidates. Select the version intended for the final cut and keep a note of the decision. Generation information or prompt restoration can bring a historical video's settings back into the editor for further work.

Editing a prompt does not alter existing footage. A new generation creates a new result. Keep useful candidates and avoid clearing history before final export.

Process generated video

Available tools can include frame capture, subtitle removal, upscaling, download, library saving and precise editing. Captured frames can become the next shot's opening frame or blocking reference. The source feature description gives subtitle removal a 24-hour post-generation availability window; current access and applicability are determined by the studio.

After generation, use available frame capture, subtitle removal and upscaling tools.

After generation, use available frame capture, subtitle removal and upscaling tools.

CoolShowAI

Continue with continuity, precise video editing or review and export.

Candidates, batch models and saved edits

• A shot can generate up to 4 candidates per submission. Compare action, lip sync and continuity before choosing the adopted version.
• You can change the model for every shot in an episode. Check all durations and references first, especially when moving from 2.5 to 2.0.
• Image and video prompt editors use autosave. Confirm the displayed save state before changing shots, particularly on an unstable connection.
• Seedance 2.5 shot prompts support 15,000 characters. The older 3,000-character record is not the current limit for that model.
• AI conversation can refine a shot prompt over several turns. Request a specific change, such as retaining dialogue while making the first shot a locked medium shot, then review the full prompt and references.
• In older projects with broken associations, use Update prompt (提示词更新) where offered to rebuild links. Visible names alone do not prove valid references.

Subtitle removal, upscaling and failed tasks

Choose Remove subtitles (字幕擦除) at the video's upper-right. The processed version carries a marker. The documented window is within 24 hours of generation, so process selected clips promptly.

Quality enhancement offers output tiers such as 720p, 1080p, 2K and 4K, depending on the entry. Upscaling does not correct wrong faces, actions or props. Failed video tasks can be removed; copy the shot ID and exact error before deleting evidence needed for troubleshooting.

Environment-only and prop-only shots can omit characters. Remove irrelevant identity references to avoid introducing people into an empty scene.