NeuroAI NEUROAINEUROAI.SITE
ESC

The editor behind a billion TikTok clips just became an AI studio with 85 built-in tools

ByteDance's CapCut — called JianYing (剪映) in China — says it has passed 660 million users and now bundles more than 85 AI features, from avatars to script-to-video, into one editing surface.

2026-10-18 · 929 words · NeuroAI
The editor behind a billion TikTok clips just became an AI studio with 85 built-in tools

For years the app was the last step before you hit "post": trim the clip, drop in a caption, ship it. Now the same app wants to be the first step too. At a creator event in Singapore in September 2026, ByteDance's CapCut — known domestically as JianYing (剪映) — announced it had crossed 660 million users across more than 170 markets and 30 languages, and that it now ships more than 85 AI tools inside the product.

The framing shifted from "editor" to "AI creation platform." The numbers suggest why the company cares.

From finish line to starting block

CapCut began in 2019 as JianYing (剪映) in China and expanded internationally as CapCut in 2020, riding the TikTok wave as the default lightweight editor for short video. Its growth was already staggering: CNET reported it had reached roughly 200 million active users by 2023 and earned about US$100 million that year, while Chinese business outlets put the combined JianYing–CapCut user base at over 800 million monthly actives by the end of 2024, with revenue nearing RMB 10 billion (≈ US$1.4 billion / HK$11 billion) and triple-digit growth.

The 2026 milestone reframes scale as an AI story. According to company figures shared at the Singapore event and reported by multiple tech outlets, CapCut users generated more than 6 billion image, video, and effect assets during 2026, with over 7 million creators using its AI features every day.

What the 85 tools actually do

The AI layer is not one button; it is a stack of discrete capabilities layered across the editing pipeline:

  • Auto-captions and translation — transcribe speech and drop styled subtitles, with auto-translate across many languages.
  • Text-to-speech and voice cloning — turn scripts into narrated voiceovers, or clone a voice for consistency.
  • AI avatars — synthetic presenters that can recite a script, useful for training and marketing videos.
  • Script-to-video — assemble scenes from a written prompt rather than from footage.
  • Background removal, upscaling, noise reduction, smart cut — the unglamorous but high-frequency edits that used to take a human.

Crucially, the generation side is fed by ByteDance's own models: the event highlighted integrations with Dreamina (即梦) Seedance for video, Dola Seedream for images, and Seed Audio for sound. In other words, CapCut is becoming the consumer face of ByteDance's entire generative stack — the place where the models meet a publish button.

The "Agent" turn

Two newer pieces show where the product is heading. AI Lab lets a creator brainstorm in plain language; Infinite Canvas turns that into characters and scenes; and an in-editor assistant called Edit Pilot applies edits based on intent. A separate mobile assistant, Xiaoying (小映), wraps idea generation, multi-track editing, and a merchant-marketing mode into one chat-style surface.

To monetize the shift, JianYing introduced an AI Ultra membership (会员) that folds AI credits into its pro editing plan at roughly RMB 28 per month (≈ US$4 / HK$30), and said it would invest about RMB 150 million per year (≈ US$21 million / HK$165 million) in creator incentives. That is ByteDance's first systematic attempt to charge consumers for AI tooling at scale.

Labeling the machine-made stuff

One detail matters for trust. CapCut said it has built invisible watermarks and adopted C2PA Content Credentials — an industry standard for signaling that media was AI-generated — across its AI creation tools. For a product that sits one click from a billion feeds, a provenance tag is less a nicety than a requirement, and it aligns CapCut with the broader global push to label synthetic media rather than hide it.

A resilience footnote

CapCut's availability has not been perfectly smooth. In January 2025 it was briefly pulled from U.S. app stores alongside TikTok under U.S. legislation, then returned days later. The episode was a reminder that a creator tool this embedded still depends on regulatory weather it cannot control — a risk worth naming even as the user base keeps climbing.

What readers can do now

  • Use the free tier to test the AI stack. The free editor plus limited AI credits is enough to try auto-captions, script-to-video, and an avatar without paying — a low-risk way to see where the tools help and where they still need a human.
  • Adopt the watermark and C2PA settings. If you publish AI-generated clips, turn on content credentials so platforms and viewers can tell what was generated; it protects your own credibility.
  • Watch the membership math. If you are a daily creator, compare the AI Ultra credit allowance against your real weekly usage before subscribing; the value flips quickly once you lean on generation features.

Honest limitations

The user and usage figures here — 660 million users, 85 AI tools, 6 billion generations, 7 million daily AI users — are company claims reported by tech outlets (Tech Edition, Tech Coffee House, and others) at the Singapore event; I have not independently audited them, and "users" may mean registered accounts rather than active ones. The older 2023–2024 numbers (200M actives, ~US$100M revenue, 800M combined MAU, ~RMB 10 billion revenue) come from CNET and Chinese business reporting and are consistent across sources but still vendor- or estimate-based.

I have not tested CapCut's current feature set hands-on, so quality claims about Edit Pilot or Xiaoying are described from reporting, not personal use. Two dimensions are uncovered: the precise training data and moderation behavior of the integrated Seedance/Seedream/Seed Audio models are not detailed in the consumer materials, and the long-term effect of the 2025 U.S. app-store disruption on CapCut's Western user base is not quantifiable from the figures released since.

Related coverage

More in “Generative Media” → · Back to home · Markdown version