CyberVerse
Lynpoint/CyberVerseReviewed by AI
Checked by RepoPicks Radar on
What is CyberVerse?
CyberVerse runs on your own machine or rented server and gives you an AI character you speak with in real time through the browser, with interruptions like a video call. It keeps each character's memory and can read your documents to answer in character. Voice-only mode needs no graphics card, just paid voice and model services. The talking face is optional and heavy: the README's real-time table uses an RTX 5090 or RTX PRO 6000, or a cloud avatar service. Setting it up means Node, Go, Conda, Python, FFmpeg and three terminals.
How easy is it to install CyberVerse?
E · developers only · Linux
Linux is the only documented system and it is for developers: the prerequisites are "Node 18+, Go 1.25 (required: protoc-gen-go, protoc-gen-go-grpc), Conda, Python 3.10+, FFmpeg, libopus-dev, libopusfile-dev, libsoxr-dev, pkg-config", then you clone, edit two config files, run make setup and keep three terminals running. The talking face adds CUDA 12.8, a specific PyTorch build, model weights from Hugging Face and, for the optional speed-ups, compiling kernels from source.
Coming soon: install in 1 click with the RepoPicks app.
CyberVerse on GitHub
- Stars
- 1,945on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +195stars in 6 days
Stars over time
Facts
- Runs on
- Linux
- You need to install
- Python, Node.js, Go, NVIDIA graphics card, FFmpeg
- To build from source
- Python, Node.js, NVIDIA graphics card, FFmpeg
- Disk space
- 410 MB–1.3 GB · estimate
- Accounts or keys
- A model and voice provider (Alibaba Cloud DashScope/Qwen, Volcengine Doubao, Atlas Cloud, Cheaper Inference or another OpenAI-compatible service), Baidu Xiling Digital Human, Xunfei Digital Human, Vidu S1 (cloud avatar)
- Cost
- Free, API costs extra · The code is free (GPL-3.0). The README says "Runtime cost depends on the realtime voice/omni/LLM/TTS/ASR providers you configure", so you pay per use for the model, speech and avatar services, or for the GPU that runs the local avatar models.
- License
- GPL-3.0 (Copyleft)
- Can replace
- D-ID, HeyGen, Character.AI
- Last change on GitHub
- Oct 21, 2026
CyberVerse risk signals (6)
- Attentionmakes a real persons face speak
The README offers to animate a real person from a single photo, including someone who has died: "Want to see someone you've lost again, hear their voice, watch them smile at you?" and "Just one photo. CyberVerse makes them alive." The demo characters include a figure from a novel (Xiaolongnü), and the README notes they "are not bundled with CyberVerse and are not provided for commercial use". Using another person's face or voice, or a character someone else owns, needs their permission, and several countries have rules about images of a person made without consent.
AI review
- Infoneeds a top end graphics card
The real-time table lists "FlashHead 1.3B | Pro | RTX 5090 | 2 | 512×512 | 25+" and "LiveAct 18B | RTX PRO 6000". The README adds "For pure voice sessions, no local avatar GPU is required", so the talking face is the part that needs the hardware or a cloud avatar service.
AI review
- Inforeferral links in readme
Several links in the README carry the author's codes: the sponsor block says "Register through this invitation link" to "passport.compshare.cn/register?referral_code=IBmJcGPVu1RF78dMihkQCX", the quick-start cloud image repeats the same code, and the Vidu S1 row links to "vidu.cn/subscription/personal?bonus-code=GWCK5M".
AI review
- Infoweights and services from outside
The avatar models are downloaded separately from Hugging Face or ModelScope ("hf download Soul-AILab/SoulX-FlashHead-1_3B", "hf download Soul-AILab/LiveAct"), with a mirror suggested for mainland China ("export HF_ENDPOINT=https://hf-mirror.com"), and optional speed-ups are built from other repositories (LightX2V, CUTLASS, SageAttention, FlashAttention).
AI review
- Infoconversations go to cloud providers
The default setup sends speech and text to outside services: config/env asks for "DASHSCOPE_API_KEY" (Alibaba Cloud), "DOUBAO_ACCESS_TOKEN"/"DOUBAO_APP_ID" (Volcengine), "ATLASCLOUD_API_KEY" or "CHEAPER_INFERENCE_API_KEY", and the cloud avatars use "BAIDU_XILING_APP_ID"/"BAIDU_XILING_APP_KEY". Character history and imported documents are kept on local disk.
AI review
- Infovery young project
Created on 2026-04-18 with a single release, v0.1.0 from 2026-05-16, and pyproject.toml still says version = "0.1.0"; the roadmap and part of the community live in Chinese (Yuque and a WeChat group).
AI review
Evidence is shown in its original language.
What Radar could not check about CyberVerse
- Whether the real-time avatar keeps up on a card weaker than the ones in the README
- How much the voice, model and cloud avatar services cost per hour of conversation
- Whether the full setup completes on a current Linux without manual fixes
- Real disk use after the avatar model weights are downloaded
- Whether the digital human looks and sounds as in the demo videos