← Back to products
Real-time Digital Human · End-to-End

Yanlin Digital Human

An end-to-end real-time digital human. Hand it a deck and it presents slide by slide, answers questions automatically, and takes interruptions mid-flow — the always-on presenter for service halls, showrooms, and briefings.

Contact for a trial → Watch the Q&A demo See what it does
Yanlin character
Yanlin · Live Q&A demo02:24 · Original Chinese audio
The presenter stage as it runs: the deck advances on script at the left, the avatar streams live at the right, and the input box takes a question at any point.
<300 msFirst audio · question to speech
≈1 sFirst frame · lip-sync begins
25 FPSReal-time render · streamed throughout
On-premisesData stays inside your network

Yanlin is Coolwei's real-time digital human presenter. Hand it a deck and a script and it presents page by page; when the audience asks a question, it finishes the current sentence, answers, and carries on — speaking and appearing on screen in real time throughout.

Yanlin presenting slides page by page

Presents the deck, slide by slide

Give Yanlin the script and the slides. It presents page by page and emphasizes what matters — briefings, showroom tours, and standing duty in service halls. Updated scripts take effect immediately.

Demo recording frame: Yanlin avatar and live answer panel

On duty, around the clock

Service halls, showrooms, briefings, and corporate training — Yanlin holds the post for the long run, with no limits on hours or headcount. Updated scripts take effect immediately, the same presenter is reused indefinitely, and on-premises deployment keeps the data inside your network.

Auto Q&A: how it speaks in real time

Visitors ask at any time. Yanlin finishes the current sentence, answers, and picks the deck back up — speaking and appearing on screen live, with no wait on a text reply. What makes that possible is a shorter chain.

The usual approach splits one sentence into three passes: turn the speech into text, hand the text to a language model, then synthesise the answer back into speech. Three stages in series, each waiting on the one before, and the delay accumulates at every step. Yanlin takes a different route — a single omni model hears the question and speaks directly, with no text relay in between.

Same question, two chains Typical cascaded chain Yanlin · end-to-end Question Speech→text Text model Text→speech Lip drive On screen Question Omni model speech in, speech out — no text relay Lip drive On screen Retrieval Deck · knowledge base Matching passages only Three conversions gone 6 stages · each waits its turn 4 stages · one pass straight to speech Measured under 300 ms to first audio
Chain structure only — no time axis. The latency figures are live readings from the presenter stage (273 ms to first audio, 988 ms to first frame) and vary with script length and deployment environment.

The question does not go straight to the model. The system first retrieves the relevant passages from the deck and your knowledge base and hands only those over — so answers stay anchored to your own material, and when the deck or the knowledge base changes, the answers change with it. No retraining.

The avatar and the presentation scheduler hang off the same chain: the lips track the speech in real time and the video streams out throughout. When someone interrupts, the scheduler finishes the current sentence, answers, and picks the deck back up. Each layer can be deployed on its own.

The six layers under the stage 01 Ingest Deck + script Knowledge base Typed or spoken 02 Retrieval Live lookup per question Matching passages only 03 Model Omni model Hears the question Answers aloud 04 Avatar Lips and expression track the speech 05 Delivery Real-time stream Interrupt · answer · resume 06 Deployment On-premises Data stays in your network
Grouped by capability, so each layer can be placed according to your compliance and compute constraints.

Yanlin is now open for trials

Built for government service halls, showrooms, executive briefings, and corporate training, with on-premises deployment supported. Book a demo and put Yanlin to work on your real scripts.

Contact the business team →