CapCut offers two AI video workflows. One turns a text prompt or a reference image into a single generated clip. The other turns a topic or script into a full multi-scene video with voiceover, subtitles and music. Pick the route first; the models, limits and costs follow from it.
I’d generate one short, well-defined shot before committing the rest of the edit to it. A clean thumbnail can hide unstable motion or changing product details. Keep the brief small, inspect the whole saved clip and use only the part that holds together.
The short answer on cost: Seedance 2.5 runs on credits and is billed by the second for Pro and Ultra members, according to CapCut’s pricing explainer for PC. CapCut’s own pages disagree on several details, including free access, maximum resolution and regional availability. I flag each conflict where it matters and name the control that settles it.
Use one generated clip for a shot, cutaway, product beat or transition. Use the full-video route when you need a narrated, multi-scene draft.
Seedance 2.5 is billed by the second, so duration changes the cost. Check pricing and your balance before you generate.
CapCut’s pages list different models, lengths and resolutions. The selector in your account is the one that counts.
A clean opening frame doesn’t rescue hands, text or continuity that break later in the shot.
Start with the right CapCut generator
CapCut uses “AI video generator,” “AI video maker,” “AI media” and “Video Studio” for tools that sound alike. Underneath, they do two different jobs.
Video clip
- One prompt or reference-driven shot
- Model-specific duration, resolution and ratio controls
- A better fit for B-roll, a hook or a single visual beat
- Desktop routes: AI Video on the homepage or AI media in the editor
Full video
- A multi-scene draft built from a topic or script
- Style, voiceover, avatar, subtitle and music controls
- Best treated as a storyboard-to-draft workflow
- Covered in our CapCut Video Studio guide
Pick by the output you need, not by the largest heading on the landing page. Our AI workflow map separates the neighboring routes by input and output.
How to generate one AI video clip in CapCut Desktop
CapCut documents two Desktop routes to a single generated shot. They name different models, so check which one your build shows before you follow a tutorial.
Homepage AI Video with Seedance 2.5
CapCut’s Seedance 2.5 Desktop instructions start from the homepage, not from inside a project.
- Open AI Video. On the CapCut Desktop homepage, choose AI Video, then Text to video, Image to video or Omni reference. CapCut's Desktop page lists Omni reference for version 840 or later and calls the other two modes available on “all versions” without defining that scope. Check which modes your installed build actually exposes.
- Add references. Click + to import images or videos you have the rights to use. A video longer than the supported duration opens in a lightweight trimmer, where you cut it to 15 seconds or less.
- Write one shot. Describe one subject, one visible action, the setting, one camera move and anything that must stay stable. Use an @ reference to tell the model which file to follow.
- Set model, resolution and ratio. CapCut lists Seedance 2.5, 480p or 720p with 1080p marked as coming soon, and ratios from Fit and Original to 9:16, 1:1, 3:4, 16:9, 4:3 and 21:9.
- Check the cost, then generate. Confirm your balance and the current pricing for the seconds you selected. Generate, then move only the clean result into the edit.
AI media in the editor with VEO 3.1 or Sora 2
The CapCut AI video tool page describes a second route inside an open project.
- Open AI media. In the editing workspace, choose AI media from the left panel.
- Choose the input. Pick Text to video for a written shot or Image to video when you have an authorized reference. CapCut says image to video can use a first and a last frame.
- Pick the model. Use Model to choose VEO 3.1 or Sora 2, then set the duration and aspect ratio. The page says text-to-video clips run up to 8 seconds.
- Generate and inspect. Check motion, identity, text, logos and audio across the whole clip before it goes onto the timeline.
- Export a proof. The export panel’s resolution (up to 8K), frame rate and bitrate apply to the finished project, not to the generated shot.
Make a full AI video from a topic or script
This is the route Google’s AI answer summarizes and CapCut’s AI video generator page walks through. Instead of one shot, you get a draft with scenes, voice and captions.
CapCut Web
- Choose AI video maker > Instant AI video
- Pick a style, then paste a script or use Generate script
- Set the aspect ratio, voiceover and duration, then click Create
- Add music, subtitles or effects and export
CapCut Desktop
- Choose AI video maker on the homepage, then New project
- Enter a topic, paste a script or use Create one with AI
- Set the ratio and style, then click Create
- Adjust voiceover, subtitles and music in the Scenes panel, then export
CapCut’s help center says the Desktop AI Video Maker is still in a testing phase, rolling out gradually and possibly only in some regions. If it isn’t on your homepage, update the app, sign out and back in, or use the Web version. For narration and ad-style scripts, the CapCut Script to Video workflow covers the script side in more depth.
Models and limits CapCut documents
CapCut doesn’t publish one model table. The figures below come from four of its own pages. They don’t all agree, and they don’t all describe the same route.
| Route | Models named | Length | Resolution and ratio | Inputs |
|---|---|---|---|---|
| Desktop homepage > AI Video | Seedance 2.5 | Up to 30 sec per generation; reference video trimmed to 15 sec or less | 480p or 720p, 1080p coming soon; Fit, Original, 9:16, 1:1, 3:4, 16:9, 4:3, 21:9 | Text, image, Omni reference (v840+) |
| Editor > AI media | VEO 3.1, Sora 2 | Text to video up to 8 sec | Duration and ratio selectable; 8K is the export setting | Text, or an image with first and last frames |
| AI video maker (full video) | Page headline: Seedance 2.5 | Video duration setting | Aspect ratio choice; export described as HD | Topic or script, style, voiceover |
| Seedance 2.5 feature page | Seedance 2.5 | 30 sec native; 180 sec in beta Long Video Mode | Native 4K claimed | Up to 50 multimodal references |
Two conflicts and one route difference matter. The Seedance 2.5 feature page advertises native 4K, while the Desktop setup steps stop at 720p with 1080p still coming. The Seedance 2.5 tool page promises up to 30 seconds per pass; the separate AI media route lists up to 8 seconds for VEO 3.1 and Sora 2 text to video. Those are different models and routes. And the generator page calls Seedance 2.5 available globally, while the feature page describes a regional rollout for subscriber accounts. If 30 seconds or a higher resolution is missing from your selector, a prompt won’t unlock it.
What CapCut AI video costs
CapCut PC is free to download, but AI video generation isn’t uniformly free. CapCut’s Seedance 2.5 launch announcement for PC says Pro and Ultra members use the model through credit consumption, billed by the second. The same announcement mentions a limited-time Pro discount on AI Video, AI Edit and AI Extend during a two-week launch period, and points readers to the in-product pricing page for current rates.
One CapCut page muddies this. The Seedance 2.5 feature page says you can start with free daily credits and that Pro unlocks 4K and the 180-second mode. Check whether the daily allowance appears in your account, then read the paywall, available balance and current rate for the selected duration.
The credit mechanics come from CapCut’s Credits Rules and its AI troubleshooting help:
- Three types: subscription, activity and purchased credits, spent in that order.
- Expiry: the Credit Rules use a subscription month, while credit-types Help uses the subscription billing cycle; check your balance's displayed expiry. Purchased credits last two years under the Credit Rules; activity credits follow their award rules.
- No way out: credits can’t be refunded, cashed out, transferred to another account or used to buy a plan.
- One account, every device: purchased credits work on any device signed in to the same account.
- Where to see the balance: on Desktop, Join Pro > Use credits for Pro AI features; on Web, your profile > Credits.
Because Seedance 2.5 is billed by the second, a 30-second take uses more credits than a 5-second take on the same model. Settle the model, duration, resolution and ratio first, then check pricing and your balance before you generate. Our CapCut AI credits guide goes deeper on balances and expiry, and the CapCut free vs Pro breakdown covers what each plan includes.
How to choose a model without inventing a winner
Use CapCut’s model descriptions to choose a starting point: longer action, synchronized sound or a shorter text-led shot. Then compare results on the same prompt and settings, checking motion, identity and usable seconds before spending on a longer take.
CapCut documents up to 30 seconds per pass, timestamped prompts and multiple references. Use it when the action needs room to develop.
CapCut credits it with audio-video synchronization, stable motion and better handling of complex or vague prompts, within a shorter documented clip length.
CapCut describes livelier audio, more accurate text-to-video and refined voiceovers. The AI media page gives it the same 8-second ceiling.
CapCut still has pages for Dreamina Seedance 2.0 Mini, Seedance 1.5 Pro and Seedance 1.0. If your picker shows them, compare cost and duration on identical settings.
For a paid comparison, keep the prompt, references, ratio, duration and resolution identical. Change one model at a time. Otherwise you’re comparing different assignments, not different models.
Write the shot before you spend
“Make a cinematic product video” leaves almost every decision to the model. A stronger prompt gives it one clear shot to solve: subject, action, setting, camera movement and hard constraints. CapCut’s guide to 30-second Seedance 2.5 videos makes the same case: one scene, one action, one visual mood, with the duration and format decided before you generate.
[Subject] [does one visible action] in [specific setting and light]. Camera: [framing and one movement]. Keep [identity, product shape or composition] consistent. No [unwanted text, logos, extra limbs or abrupt cuts].
Seedance 2.5 also accepts timing cues. CapCut’s Seedance 2.5 page shows the pattern with a character who sits at 00:06, turns at 00:12 and exits the frame at 00:24. Use timestamps only for actions that must land on a specific second; every extra cue is one more thing the model can miss.
- Choose the destination first. A vertical Reel and a 16:9 YouTube insert need different framing before generation.
- Ask for one camera move. “Slow push-in” is easier to judge than a prompt requesting orbit, crane, zoom and handheld motion at once.
- Name what must stay stable. For a product, that may be label shape and color. For a character, it may be clothing, hairstyle and facial identity.
- Avoid generated on-screen copy when spelling matters. Leave clean composition space and add final text in the editor.
- Use reference media only when you hold the rights. A clearer reference may improve control, but it also becomes uploaded input governed by the service terms.
How to review a generated clip
Watch once at normal speed, once muted and once frame by frame. The muted pass exposes motion and continuity problems that music can hide. The frame-by-frame pass catches a logo, hand or face that changes for only a fraction of a second. CapCut’s 30-second guide adds two habits worth keeping: generate a few takes and compare them, and review at desktop size rather than approving on a small screen.
- Prompt adherence: did the requested subject, action and setting appear, or did the model quietly replace part of the brief?
- Motion coherence: do limbs, wheels, fabric, reflections and background objects move in a physically readable way?
- Temporal continuity: do identity, clothing, product shape, lighting and object count stay stable from first frame to last?
- Camera behavior: is the movement the one you requested, without an unexplained cut, zoom or drift?
- Audio: when a model returns sound, check speech sync, noise, clipping and whether the track is usable rather than merely present.
- Usable seconds: mark the clean in and out points. A nominal eight-second clip may contain fewer seconds you would publish.
When trimming, masking or replacing audio can hide one local defect, fix it in the edit. If identity, geometry or the requested action fails across the shot, a rerun with a narrower prompt is usually cleaner than building a repair stack around unstable frames.
Why CapCut’s product page may not match your panel
CapCut’s pages describe different routes, plans and rollout stages, so their numbers clash. Read every headline figure against the route it belongs to:
- Length: 8 seconds for VEO 3.1 and Sora 2 text to video in AI media; up to 30 seconds per Seedance 2.5 pass; 180 seconds only in Seedance’s beta Long Video Mode.
- Resolution: “native 4K” on the feature page; 480p and 720p, with 1080p coming soon, in the Desktop setup steps.
- Export vs generation: “up to 8K” is the editor’s export setting. It doesn’t make the generated frames 8K.
- Availability: “available globally” on one page, a regional rollout to subscriber accounts on another, and a Desktop AI Video Maker still in testing.
The generated resolution can be lower than the project you export. Dropping a 720p clip onto a 4K timeline won’t add the missing detail. Check the resolution beside the selected model, then inspect the exported file’s properties.
Free use, watermarks and commercial rights
Can you use CapCut’s AI video generator for free? Partly. The app is free to download, and CapCut’s generator page says the full-video route exports in HD for free. Seedance 2.5 clip generation is described as credit-based for Pro and Ultra members. Check any free allowance, the balance and the paywall shown in your own account before planning a project around free output.
CapCut’s watermark help says most exports are watermark-free, but a template ending clip, a stock asset that needs attribution or the default outro can add one. Inspect the generated clip and the final export; don’t promise a client “no watermark” before you’ve watched the file. If a mark does appear, our guide to removing the CapCut watermark covers the legitimate fixes.
CapCut’s AI video page calls VEO 3.1 and Sora 2 output “copyright-safe” for commercial projects, and the Seedance 2.5 feature page allows commercial use in accordance with CapCut’s terms. Read the CapCut Terms of Service before relying on either line. They say you must have all necessary rights, licenses and permissions for the inputs you give third-party AI technologies, including publicity clearances, and they prohibit using outputs in ways that violate other people’s rights. Third-party provider terms can apply as well. Use media you own or can license, get consent from identifiable people, and keep confidential client footage out unless your agreement allows the upload.
Fix the problem before paying for another reroll
CapCut’s AI troubleshooting help says this message means your region isn’t supported for AI services yet, for example because of local regulations. That makes it a location block, not a prompt problem.
The first means no usable credits; the second means the tool needs an active Pro plan even if you have credits. If you’ve paid and still see it, work through our CapCut Pro not working fixes.
CapCut says the feature is in a testing phase and rolling out gradually, possibly by region. Update the app, sign out and back in, or switch to CapCut Web.
CapCut notes that Desktop often fails silently. Try the same feature on CapCut Web; if it works there, your Desktop app may need updating through Help > Check for Updates.
Omni reference needs version 840 or later. Files over the creative limit can’t be imported, and long videos must be trimmed to 15 seconds or less.
Use our CapCut not working checks for sign-in, cache, network and app-build problems.
CapCut AI video generator FAQ
How do I make an AI video in CapCut?
For one generated clip on Desktop, open AI Video on the homepage (Seedance 2.5) or AI media inside a project (VEO 3.1 or Sora 2). Describe one shot, set the model, duration and ratio, then generate. For a full narrated video, use AI video maker > Instant AI video on Web, or AI video maker on the Desktop homepage.
Is the CapCut AI video generator free?
Partly. CapCut is free to download, and its generator page advertises free HD export for the full-video route. CapCut also says Seedance 2.5 is used by Pro and Ultra members through credits billed by the second, while one feature page mentions free daily credits. Check whether that allowance is available in your account, along with the balance and cost shown for the selected settings.
How much does the CapCut AI video generator cost?
Seedance 2.5 consumes credits billed by the second, so duration drives the cost. Set the model and output options, then check the in-product pricing page for current rates. Purchased credits last two years and can’t be refunded.
Which AI video models does CapCut offer?
CapCut’s pages name Seedance 2.5 for the homepage AI Video route and VEO 3.1 and Sora 2 for the editor’s AI media panel. Separate CapCut pages also cover Dreamina Seedance 2.0 Mini, Seedance 1.5 Pro and Seedance 1.0. Rollouts vary by region and account, so your model picker is the current list.
What is the maximum CapCut AI video length?
CapCut documents up to 30 seconds per Seedance 2.5 generation and up to 8 seconds for VEO 3.1 and Sora 2 text to video. Its Seedance 2.5 feature page also describes a 180-second beta Long Video Mode. The full-video route has its own duration setting.
Can CapCut generate AI video in 4K or 8K?
The Seedance 2.5 feature page claims native 4K, but CapCut’s Desktop setup steps list 480p and 720p, with 1080p coming soon. The up-to-8K figure is the editor’s export setting. Exporting at a higher resolution doesn’t add detail to a lower-resolution clip.
Does CapCut AI video have a watermark?
Most CapCut exports are watermark-free, according to CapCut’s help center. A template ending clip, a stock asset that needs attribution or the default outro can add one. Inspect both the generated clip and the final exported file.
Can I use a CapCut AI video commercially?
CapCut’s product pages say yes, but its Terms of Service make you responsible for having the rights to your inputs and bar using outputs in ways that violate others’ rights. Third-party model terms can apply too. Check people, logos, music and text before commercial use. This is a rights check, not legal advice.
Is CapCut a good AI video generator?
It’s strongest when you want generation next to a full editor: the clip lands on the same timeline as your captions, audio and export. The weak spots are pages that disagree on models, limits and free access, and per-second credit billing that makes long rerolls cost more. Judge it by your own picker and pricing, not the product page.
Verdict: approve the shot, not the product-page promise
Use the clip route when
- You need one specific visual beat
- Your picker exposes the ratio and duration you need
- You can afford at least one controlled rerun
- You will inspect the full clip before publishing
Use another workflow when
- You need a narrated, multi-scene draft
- Exact typography or product geometry must survive untouched
- Rights prevent uploading the reference media
- The cost of the seconds you need doesn’t justify a speculative result
CapCut’s AI video generator is easiest to use when you keep the assignment small: one shot, one camera idea, one destination and a cost you’ve checked before generating. Save the output, count the usable seconds, and move only the clean part into the edit. That’s slower than believing the demo reel, and much cheaper than building a project around a clip you never properly checked.