Bilibili HAI × BibiGPT
On 2026-05-11 Bilibili confirmed it is in the summer-launch window for Codename H / HAI — an AI video-podcast tool that compiles a 1,000-word script (text or audio) into a 6-minute Bilibili video and is targeting a 3-minute output cycle at general availability. BibiGPT pairs with HAI on the front end: extract transcripts from any Bilibili / YouTube / podcast source, then push the cleaned script into HAI for the auto-edit step.
Key facts (90-second read)
As of 2026-05-12: Bilibili confirmed Codename H / HAI is in its summer-2026 launch window (ainvest.com 2026-05-11). HAI is a text-or-audio-to-video AI tool that turns a 1,000-word script into a roughly 6-minute Bilibili short, with a 3-minute end-to-end target at general availability. It is positioned against CapCut AI and ListenHub's video-overview feature, distributed natively into Bilibili's creator pipeline. There is no public sign-up surface yet. For Bilibili creators, the practical play today is to fix the upstream half — get clean, chaptered transcripts out of existing long-form content — so the eventual HAI input is as good as it can be.
Features
What Bilibili confirmed about HAI / Codename H
Bilibili's strategic disclosures (May 2026) describe HAI as a text-and-audio-to-video assistant that drops scripts straight into a finished short clip — built to keep up with the platform's pivot from long-form to short-form on the home feed.
1,000-word script → 6-minute clip (current target: 3 minutes)
Bilibili executives have benchmarked HAI on a 1,000-Chinese-word input script producing a roughly 6-minute video at the current internal build. General-availability target is a 3-minute end-to-end cycle.
Text or audio input, multimodal output
HAI accepts either a written script or a recorded audio narration. The output is a Bilibili-native short video with auto-selected B-roll, captions and music — the same surface area as existing competitors like CapCut AI and ListenHub's video overview mode.
Summer 2026 launch window
As reported on 2026-05-11 (ainvest.com), HAI is in its summer-2026 release runway. Bilibili has not committed to a specific GA date but flagged Codename H as one of its flagship AI products for China's content market.
What HAI means for Bilibili creators (and where BibiGPT fits)
Auto-edit pipelines turn the script into a clip, but they need a good script to start with. BibiGPT covers the half that HAI doesn't — pulling that script out of existing video, podcast and livestream sources.
BibiGPT extracts the script, HAI assembles the clip
Drop a Bilibili / YouTube / podcast URL into BibiGPT to get a clean transcript with chapter markers and a rewrite-ready outline. Hand that text to HAI when public access opens — the auto-edit step takes a structured input, not a raw transcript.
Multilingual reach beyond Bilibili
HAI publishes natively to Bilibili. BibiGPT supports zh / en / ja / ko / zh-TW and exports to TikTok, Xiaohongshu, Notion, Obsidian and Logseq, so the same script can fan out to international and note-taking surfaces.
Re-edit existing content as auto-cuts
A creator's existing long-form Bilibili archive — interviews, podcasts, livestream replays — becomes raw material for HAI. BibiGPT turns each long episode into a 1,000-word distilled script ready for HAI's short-form auto-cut.
5 key facts (90-second read)
The Codename H / HAI launch in 5 facts.
- 1
Confirmed 2026-05-11
Bilibili's strategic disclosure published 2026-05-11 (ainvest.com) confirmed HAI as a flagship AI product entering its summer launch runway. Until then, Codename H was only an internal reference.
- 2
1,000 words → 6 minutes today
Bilibili executives benchmarked HAI on a 1,000-Chinese-word input script producing a roughly 6-minute video at the current internal build. The end-to-end pipeline still feels slow for short-form rhythm.
- 3
Targeting 3 minutes at GA
The internal target at general availability is to compress that pipeline to 3 minutes end-to-end — fast enough to compete with manual short-form workflows like CapCut.
- 4
Text or audio input
HAI accepts either a typed script or recorded narration. The output is a Bilibili-native short video with auto-selected B-roll, captions and music — same surface area as competing AI video tools.
- 5
Bilibili distribution as the moat
Unlike global tools (CapCut, Runway, OpusClip), HAI publishes natively into Bilibili's creator pipeline. The strategic bet is automation × distribution: tools shipped inside Bilibili are stickier than third-party uploads.
3 typical scenarios for Bilibili creators
What to do today, before HAI opens public access.
Long-form archive → short-form auto-cut feed
Use BibiGPT to extract clean, chaptered transcripts from your existing long-form Bilibili episodes, podcast interviews, or livestream replays. Distill each one into a ~1,000-word script. When HAI opens, that script is the exact input format — no re-prep needed.
Cross-platform fanout — Bilibili in, TikTok / Xiaohongshu out
HAI publishes inside Bilibili. BibiGPT exports the underlying script to Notion / Obsidian / Logseq and to social formats — TikTok captions, Xiaohongshu note cards, Twitter posts. The same source content fans out to international and note-taking surfaces too.
Multilingual script — zh / en / ja / ko / zh-TW
BibiGPT generates 5-locale rewrites of the same script. Push the zh version into HAI for Bilibili; reuse the en / ja / ko / zh-TW versions for YouTube Shorts, TikTok and Xiaohongshu — no second tool, no re-transcription.
FAQ'S
Frequently Asked Questions
Ask us anything!
Feed HAI a cleaner script — extract Bilibili / YouTube / podcast transcripts with BibiGPT first
Auto-edit tools only as good as the script you give them. Paste any Bilibili / YouTube / TikTok / podcast URL into BibiGPT, get an accurate transcript, chapter list and rewrite-ready outline — then route that script into HAI when the public launch lands.