ScrumSurvivor
A real-time lip-sync avatar so your GPU can sit in standups for you.
Last verified:
What is ScrumSurvivor?
ScrumSurvivor is an open-source Windows application that replaces your webcam feed with a photorealistic AI avatar of yourself, rendered in real time for mandatory camera-on meetings. It was created to address the problem of compulsory camera-on policies in corporate meetings, where employees feel forced into performing presence rather than focusing on the actual discussion.
ScrumSurvivor pricing
Pricing model: Freemium
ScrumSurvivor is completely free and open source under the MIT License. There is no free tier versus paid plan distinction—everyone gets full functionality at zero cost. There are no subscription fees, no ongoing costs, and no cloud inference charges. The only requirements are hardware (NVIDIA GPU with 4 GB VRAM minimum) and free software dependencies (Python 3.10+, OBS Studio, VB-Audio Virtual Cable) that you install yourself.
ScrumSurvivor pros
- Runs entirely locally with no cloud dependency
- No subscription fees or ongoing costs
- No data leaves your machine—privacy preserved
- Photorealistic AI avatar of yourself
- Real-time Wav2Lip lip-sync to your actual voice
- Procedural idle animations never look like loops
- Breathing animation at ~0.25 Hz for realism
- Non-repeating head sway from sum of sinusoids
- Randomized blink every 4-8 seconds
- Sensor noise adds organic camera texture
- Smoothstepped crossfades between clips—no visible cuts
- 265 ms audio delay matches video processing latency perfectly
- Cold GPU warmup prevents 18-second audio backlog
- Works on consumer NVIDIA GPUs starting at RTX 3050 Laptop
- Open source under MIT License
- Static photo input—no live webcam feed needed
- MediaPipe face detection only at startup, not per-frame
- 25 fps output at 1280×720 resolution
- Approximately 20 ms per frame on RTX 3050 Laptop GPU
- Avatar is always you—no identity impersonation
ScrumSurvivor cons
- Windows 10/11 only—no macOS or Linux support
- Requires NVIDIA GPU with CUDA support
- Minimum 4 GB VRAM requirement excludes older GPUs
- Setup takes about an hour for most users
- Requires recording your own idle clips manually
- Must install OBS Studio separately
- Must install VB-Audio Virtual Cable separately
- Requires Python 3.10+ installation and configuration
- Cold GPU startup requires warmup loop of 5-7 iterations
- Only tested on RTX 3050 Laptop GPU—other hardware unverified
Frequently asked questions about ScrumSurvivor
What is ScrumSurvivor and what problem does it solve?
ScrumSurvivor is an open-source Windows application that replaces your webcam feed with a photorealistic AI avatar of yourself, rendered in real time. It was created for people frustrated by mandatory camera-on policies in corporate meetings, where employees must perform presence by staring at grids of tired faces in bad lighting. The tool lets you attend every meeting and speak when spoken to while avoiding the compulsory performance of having a camera pointed at your actual face.
Does ScrumSurvivor send my face or voice data to the cloud?
No. Everything runs entirely locally on your machine. No face or voice data leaves your computer. There is no cloud inference, no subscription service, and no data transmission to external servers. This is a deliberate design decision to preserve privacy.
What are the minimum system requirements?
ScrumSurvivor requires Windows 10 or 11, an NVIDIA GPU with CUDA support with at least 4 GB VRAM (RTX 3050 Laptop GPU has been tested), Python 3.10 or higher, OBS Studio, and VB-Audio Virtual Cable. The tool does not work on macOS or Linux.
How does the lip-syncing work?
The tool uses Wav2Lip, which takes an 80-bin mel spectrogram window and a 96×96 face crop to produce animated facial animation with lips matching the audio. On an RTX 3050 Laptop GPU with 4 GB VRAM, this runs in about 20 ms per frame—fast enough for 25 fps with headroom. MediaPipe detects the face in a base photo once at startup, and all subsequent inference uses that fixed crop without per-frame face detection.
What happens when I'm not speaking during a meeting?
When you're silent, the pipeline plays pre-recorded short video clips of you sitting quietly at your desk, cycling through them with randomized pauses. Four procedural layers run simultaneously on top: breathing (subtle vertical oscillation at ~0.25 Hz), head sway (non-repeating micro-motion from sum of two independent sinusoids per axis), blink (fast eyelid-close/reopen every 4-8 seconds at randomized intervals), and sensor noise (per-pixel Gaussian noise simulating live camera texture). All transitions use smoothstepped crossfades with no visible cuts.
How does ScrumSurvivor keep audio and video in sync?
Wav2Lip takes approximately 20 ms of GPU time per frame, but without compensation the lips would be slightly behind the voice. The solution delays audio output by exactly the same amount as video processing latency. The AudioPresentationScheduler maintains a ring buffer of incoming microphone audio and schedules each chunk to be written to VB-Cable exactly audio_delay_ms milliseconds in the future. This gives a 265 ms window to process audio before Teams hears it, ensuring both audio and video arrive at Teams simultaneously.
What is the cold GPU startup problem and how is it solved?
The first Wav2Lip inference on a cold CUDA GPU can take 1-2 seconds because NVIDIA's driver compiles JIT kernels for your specific hardware on the first forward pass. If audio scheduling starts before kernel compilation, the audio backlog explodes to 18+ seconds permanently. The fix is a warmup loop at startup that runs up to 20 inferences using the actual face crop and monitors latency. Once inference time drops below 50 ms (typically 5-7 iterations), the pipeline opens. Any remaining backlog is discarded by a 2-second cap in the scheduler.
Is ScrumSurvivor the same as Deep Live Cam or identity swapping tools?
No. ScrumSurvivor deliberately does not do real-time face re-enactment or identity swapping. Tools like Deep Live Cam do full face replacement and require a live webcam feed as input. ScrumSurvivor uses a static photo—you take a photo of yourself, and the avatar is always you, rendered from that photo. This is the most important design decision: the tool does not impersonate anyone. It is your face, your voice, your machine, your camera output.
How do I get ScrumSurvivor and how long does setup take?
ScrumSurvivor is open source (MIT License) and available on GitHub. Setup takes about an hour for most users, mostly spent recording your idle clips and installing the required dependencies: OBS Studio and VB-Audio Virtual Cable. You need to record short video clips of yourself sitting quietly at your desk for the idle state animations.
Will people in my meetings notice I'm using ScrumSurvivor?
The author's friend has been running ScrumSurvivor for several months and nobody has said anything. The avatar is photorealistic, the lip-sync is accurate to your actual voice, and the procedural idle animations (breathing, head sway, blink, sensor noise) are designed to avoid the uncanny valley of looping video. The 25 fps output at 1280×720 through OBS Virtual Camera appears as a perfectly normal webcam to meeting software like Teams.