Unique Research · Product review · July 5, 2024
AI digital humans combine simulation and AI for human-like looks and multimodal interaction — customer service, education, entertainment, and more. With many products on the market, Unique Research compares five mainstream platforms: HeyGen, SenseTime SenseAvatar (如影), Tencent ZenVideo (智影), Silicon-based Silicon Speech (硅基硅语), and Xmov Youyan (魔珐有言).
Market & tech backdrop
iiMedia’s 2024 China Virtual Digital Human Industry White Paper: 2023 China virtual-human core market ¥20.52B, expected ¥48.06B by 2025; industry-driven market ¥333.47B (2023) → ¥640.27B (2025). Domestic stacks lean on NLP, multimodal fusion (voice/gesture/touch/gaze), and XR (AR/VR/MR). Strong scenes: 24/7 livestream, education, culture-tourism guides/museums, professional services (law, insurance, government). GenAI helps interaction quality, but answer latency still lags real-time dialogue needs — a model problem to solve. Near-term interaction leaps expected in games (NPC voice/text), livestream commerce, and functional roles (AI teachers, AI broker analysts). Rendering quality and brand-fit avatars remain commercialization gates.
1. HeyGen
Pro AI digital-human video platform: upload video → extract face/expression → avatar for hosting, support, etc. Flow: register, upload expressive face video, templates or custom; adjust size/position, voice, script, stickers/music. Voice clone (many languages; Chinese pronunciation still improving). Lab features (e.g. video translation). Pricing: monthly/annual tiers for individuals, teams, enterprises; new-user free trial time. Broad use: social, ads, education, news, CS, brand, entertainment, virtual livestream, endorsements.
2. SenseTime SenseAvatar (如影)
On SenseNova (“日日新”): text/speech/motion/image generation, NeRF; lip sync, expression, motion. Create video from text or PPT + avatar + voice; online edit background/PiP/text. Personal 2D avatar in ~5 minutes (1080P/2K/4K). Voice clone from little audio via transfer learning; multilingual. Template marketplace. Pricing: VIP tiers, custom training fees, add-on packs. Strong in short video, livestream, education, finance.
3. Tencent ZenVideo (智影)
Site: https://zenvideo.qq.com — end-to-end online edit (assets → cut → package → render → publish). Digital-human broadcast (horizontal/vertical templates, PPT import, backgrounds, Tencent Video assets, AI rewrite/expand/shrink); text-to-speech with fine pronunciation controls; article-to-video (Tencent Video licensed assets, auth required); avatar + voice custom (photo face-swap, voice clone); livestream (trial + pro with human takeover; interactive Q&A library). Pricing: custom + membership; livestream billed extra. Scenes: production, education, entertainment, news, social, livestream.
4. Silicon-based Silicon Speech (硅基智能 / 硅基硅语)
Since 2018: first commercial AIGC digital human claimed; ~500,000 cloned avatars; ~50,000 digital-human livestreams/day; valuation near $1B; ~100 patents. Custom avatar from 2–3 min script video; voice clone from corpus or online record; one-click video translation keeping timbre/style; talking photos; marketplace of models/templates/voices. Pricing in “diamonds” (¥1 = 1 diamond): e.g. general video 5/min; audio 1/min; talking photo 5/min; avatar clone S ultra-HD 8,000; livestream advanced scene clone 20,000; online record 8,000; clone E/S 980/3,980; voice clone E/S 680/3,980. Fits banking, finance, knowledge paid, science education.
5. Xmov Youyan (魔珐有言)
Site: https://youyan.xyz/platform/home/prime?from=xmovai — one-stop AIGC video: text→3D video, text→3D shots (pan/full/mid/close/CU), text→script, etc. Strong 3D naturalness; less emphasis on ultra-fast custom. Flexible monthly/annual plans. Scenes: 3D video, ads, social, education, corporate.
Five-way comparison
Voice diversity & custom: HeyGen leads (~300 voices, 50+ languages, tight lip-sync; Chinese still improving). SenseAvatar/Youyan more limited voices/custom; ZenVideo mainly CN/EN; Silicon-based stronger voice taxonomy.
Avatar & motion naturalness: HeyGen fluent body/lip + rich custom; SenseAvatar/ZenVideo more limited (SenseAvatar has 3D but scarce assets); Silicon-based average; Youyan stands out on 3D naturalness but lacks fast custom.
Production tools: HeyGen mature AI fusion; SenseAvatar covers basics (AI copy needs work); ZenVideo strong AI rewrite tools + film library + smart erase/captions; Silicon-based friendly editor (weaker AI text gen); Youyan strong 3D tooling.
Pick by scene — HeyGen for global multi-use; SenseAvatar short video/live/edu/finance; ZenVideo China content stack; Silicon-based pro finance/edu; Youyan 3D-first creatives.
Author: Jackson · Editor: Qiuping. 15 CN images/charts; attach on ingest.
— END —