Generating in the editor

Most generation in cutipAI happens on the storyboard and in the character, object and space inspectors. This page covers the rest: generating straight onto the timeline, making music and sound effects, and keeping track of every job while it runs.

Generation and automatic recovery

The price shown before generation is an estimate. New generations settle from the provider’s actual bill plus the service fee, including billed failed or canceled work. Automatic recovery includes the billed usage of every attempt. A technical failure can trigger an equivalent alternative automatically; content refusals do not. Billing may remain pending while usage is checked. Check History before retrying a job that is still being recovered.

Closing a window does not stop a submitted generation. To stop a job, cancel it explicitly in the Job Queue; provider-billed usage may still be charged.

Choose the video model

For voice replies to an uploaded recording, use Grok Voice in the Sandbox.

The top bar's Choose generation models button changes the project's Video model. Video is the only model users can change in a project. Image generation uses GPT Image 2.5, while voice, music and sound effects use ElevenLabs.

Insert a scene at the playhead

Choose Insert scene in the toolbar below the preview. Describe the new action in the textbox below the toolbar. The Before and After pictures show the two moments the new scene should connect. This tool uses H3 Max automatically and matches the whole preview picture, including visible overlays and titles. The sound it sends is every clip's own sound: muted tracks are left out, and the Mixer's faders and Solo do not change it.

Move the playhead to change the insertion point before generating. In the guide below the timeline, drag the left handle to move that point and the right handle to change the added length. You can also type exact seconds in Insert at and Add. Choose between 5 and 13 seconds; output is 768p. You need at least 1.625 seconds of playable video on each side, either inside a clip or across a cut.

Hover over Generate insert video to check its estimated cost, then generate. The submitted position stays fixed even if you move the playhead afterward. Review the complete sequence and both transitions before choosing Insert scene. Matching is a generation goal, so always check the result rather than assuming either seam will be invisible.

The new scene becomes a separate clip on a top video track. Existing unlocked tracks split at the insertion point and move forward by its actual length. Original footage is preserved, and one Undo removes the insertion. Locked tracks stay put; unlock visible tracks that would overlap the new scene. The inserted picture already contains the layers used as context.

Include audio is checked when audio was returned; clear it for a silent insert. The generated interval and complete sequence are kept in Assets. Keep in assets leaves the edit unchanged. Regenerate lets you prepare another paid generation.

If the surrounding edit changed during generation, insertion is refused and the result remains in Assets. After an interrupted job, Check finished job reads the original request without starting another generation. Review History before choosing New generation, because submitted or canceled work may still have a usage charge. Final charges use actual generation usage plus the usual fee.

Kling 4.0 in Projects

Kling 4.0 is the single Kling video choice in Projects. Choose a whole-second duration from 3 to 30 seconds and 720p or 1080p. A first frame starts the animation; an optional last frame controls its ending. Tagged characters or objects use their approved reference pictures instead. With characters and a last frame together, follow Skip the last frame to continue; the saved last frame stays in the shot.

Older Kling shot selections move to Kling 4.0 when you open the project. Existing generated videos and their original recipes stay intact. Kling no longer offers 4K generation in Projects; use another 4K model or upscale the finished video. There is no separate sound switch for Kling 4.0. Check the Estimated cost before generating.

Generate a video on the timeline

1. Choose Generate video in the timeline's tools.

2. Drag across an empty part of a video track. The length you draw is the length of the shot.

3. In Describe the shot, write the action, the camera movement and the atmosphere. Type @ to tag a character, object or space. Improve with AI polishes what you wrote — it is a small paid job.

4. Choose a Duration if the model cannot make exactly the length you drew, then click Generate, or press Ctrl+Enter (Cmd+Enter on a Mac).

The clip appears on the track, exactly where you drew it, and in the Shots folder of your assets. If the model makes a longer clip than you drew, the popover tells you how much of it is used.

If it fails, the button becomes Retry and the reason is shown. Remove draft takes the empty range away.

Generative extend, next to it in the tools, lengthens a clip you already have by generating more of it. End extensions automatically use H3 Max; beginning extensions use LTX 2.3. There is no model selector. End source context must be 1.625–60 seconds and the added footage 0.71–15 seconds; beginning context is 1–20 seconds and added footage 2–20 seconds. Hover Generate to see its Estimated cost. The original clip stays on the timeline.

Music and sound effects

1. Choose Generate → Generate audio (music/SFX) in the menu bar.

Choose Music or SFX and enter a description in the empty Prompt field. When clips are selected, Duration starts at their combined timeline lengths. Gaps add nothing; overlapping clips each count. You can adjust it manually. Long requests generate in consecutive parts.

Choose the Target track, then click Generate audio. Results start at the playhead captured when you click. Existing clips stay in place and mix with overlapping results. A track is created only when you choose New track. If the target becomes locked or disappears, completed audio stays in Assets for manual placement.

Scan the video analyzes and generates immediately on both tabs, without a confirmation sheet. Choose Selected videos only or The whole working area first; Duration applies only to text generation. Scanned audio follows the videos' original positions and preserves gaps. Long videos are processed in portions. Music analysis is a paid step and its cue estimate updates after analysis. Hover or focus either generation button to inspect its estimated cost.

Every generated SFX part passes through automatic human-vocal removal before timeline placement. The estimate includes both generation and removal. Separation can affect sound quality and may leave artifacts. A failed cleanup never places unprocessed audio automatically. Reopen this dialog to check an unfinished job or finish cleanup using the original generated sound.

Progress shows the current part and processing stage. Stop remaining jobs lets the current part finish and prevents subsequent parts from starting. Completed results and request references are kept, including when you close the dialog. Successful paid stages remain billable if a later stage fails.

Listen to all takes in Results. Use each player's seek controls and regenerate a pass or cue without scanning again. A successful regeneration replaces the same timeline clip and keeps its mix settings and earlier takes in Assets. Closing clears temporary players while submitted jobs continue. Timeline music remains on Music v2.5 and text SFX on Sound Effects v3 through the existing connection. New text SFX never switches to v2. If the connection or its verified price is unavailable, try again later. Earlier results and submitted jobs remain recoverable.

The Job Queue

If your session needs refreshing, the queue reconnects to the existing generation. If sign-in is required, open Preferences → Account and sign in to the original account; checking resumes without submitting again. A connection interruption does not prove the generation failed or was free. Check History before generating again. For an interrupted job with a request reference, use Recover result in its queue row or in History to retrieve the existing output.

Every generation you start — from the storyboard, an inspector, the timeline, the audio dialog or Auto-captions on the Text panel — waits its turn in the Job Queue. Click Job Queue in the top bar to open it.

• Each job shows what it is, how far it has got, and its Estimated cost.
• The panel's header adds up the estimate for everything in the queue.
• Drag a waiting job, or use Move up and Move down, to change the order.
• Click × to cancel a job. Cancelling is free.

Completion notices wait until the requested result is ready: video includes automatic lip sync, dialogue includes its assembled soundtrack, SFX includes voice removal and timeline placement, and images include their required compositing or repairs. Intermediate steps stay silent. With whole-queue notifications, the notice also waits for other active generation actions. If a finishing step fails, the notice reports failure; completed paid results remain available for recovery rather than being generated again. A failed job stays in the queue with its reason; if your balance ran short, Add funds takes you to your account. You are charged only for jobs that finish.

To clean existing speech, right-click an audio clip, or a video clip with its sound, and choose Denoise audio; the clip inspector's Audio section has it too. Review the original and cleaned recording before applying. Denoise cleans the clip's own recording; the Mixer's faders and Solo do not change it, and the comparison plays the recording without the clip's audio effects. The effects stay on the clip and apply to the cleaned sound. For complete audio or video recordings, use the Clean audio template.

Where results go

Everything generated in a project is kept in its assets, in folders made for you: Images (with First Frames, Last Frames and Angles inside), Shots for video, and Voice, Sfx and Music. See Organising assets.

In a 4K project (3840 × 2160), Models also offers Seedance 2.0 4K for direct 4K video generation. The model is hidden at smaller project sizes. Reducing the size switches editable selections of this model to Seedance 2.5. Generation requires verified 4K pricing to be available.