Cloud face-swap infrastructure that sits behind your MCR. < 120 ms end-to-end. 4K @ 60p. SOC 2 on the 2026 roadmap.
Facecast is a cloud-hosted face-swap service engineered for broadcast workflows. The ingest accepts NDI, RTMP, or SRT; the processed output returns on the same timecode so downstream switchers don't re-sync. Typical deployment: your camera feeds land in your MCR switcher, the face-swap pass runs on our encoder, and the program-out returns to your streaming distribution (Zoom Events, Teams Live Events, RTMP push, on-prem SRT) in under 120 ms end-to-end at 1080p60. You keep your existing router, your existing graphics, your existing failover.
Every number published here is measured, not marketed. Methodology and p99 traces available under NDA.
Facecast inserts between your source and your distribution. No new switcher, no new router, no new graphics chain.
Camera / vision mixer program-out (SDI via on-prem bridge, or NDI / SRT / RTMP direct to cloud encoder).
H.264 Baseline + H.265 Main simulcast. 50 Mbps primary, 8 Mbps mobile. Same timecode in, same timecode out.
Per-tenant model instance. Cold-start under 400 ms. Skin-tone calibration locked per session so cut between cameras doesn't drift.
Zoom Events · Teams Live Events · Webex · RTMP push · on-prem SRT. Multi-destination simulcast at no extra egress cost.
Three scenarios we shipped last quarter. Each channel strip is one real customer pipeline with the sensitive names redacted.
Fortune-500 quarterly all-hands. 14,000 simultaneous viewers across Zoom Events + internal RTMP. Face-swap applied to the host's body-double so travel constraints don't cancel the keynote. Zero dropped frames over the 47-minute program.
Subscription sports-commentary vertical. Talent swaps on a delayed feed for contract-exempt appearances. 4K simulcast to three platforms; revenue attribution per-platform on the post-stream report.
Multi-show podcast network recording 11 simultaneous video sessions. Face-swap used on anonymity-required segments (whistleblower interviews, law-enforcement sources). Post-session redaction report includes timecode ranges with face-model applied.
Three rack sizes. No per-minute billing. No credit packs. Numbers are real; your AP team will not find them negotiable inside the tier.
Real questions from the last six procurement reviews. If yours isn't here, email [email protected] — answered in one business day.
Facecast inserts as a pass-through between your switcher's program-out and your distribution chain. For on-prem SDI, our 1U bridge appliance accepts up to 12G-SDI, encodes to SRT, and pushes to our cloud encoder; the return feed comes back as NDI or SRT on the same timecode. Your existing switcher, router, and graphics chain don't change. Typical installation: half a day on-site plus a 30-minute remote calibration call.
Under 120 ms at 1080p60 (p99 over a 30-day window), measured from the source-side SDI frame leaving your switcher to the face-swapped SRT frame returning. 4K adds approximately 28 ms. These numbers are in our Rack Spec PDF and we publish the traces under NDA. If your use case requires < 80 ms (sports commentary with talent-visible earpiece feedback), we run a dedicated encoder pod — ask during the demo.
Active-active failover across three regions. The source stream is multi-cast to primary and standby encoders; the distribution side receives a single stream, with a sub-second hot-swap if the primary loses health-check. We've never missed a second of program time in 14 months of production traffic. If you require on-prem-only processing with zero cloud dependency, that's the Enterprise tier.
Source frames are held in memory only for the duration of the face-model pass (typically 40–80 ms) and discarded. On standard tiers we retain 24 h / 30 d of metadata (timecodes, face-model instance ID, latency samples) for billing and support; no video bytes. Enterprise contracts can opt for zero retention beyond the active session. DPA and GDPR Article 28 data-processing agreement available on request.
SOC 2 Type II audit completed 2026-Q2; report available under NDA. ISO 27001 certification in flight, expected 2026-Q4. CCPA and GDPR Article 28 DPAs are standard on all contracts. If your procurement requires a specific control mapping (NIST 800-53, HITRUST), send the framework and we'll return a gap analysis within 10 business days.
On Enterprise. Upload your fine-tuned weights in ONNX or TensorRT engine format; we'll containerize the inference wrapper and run it on a dedicated pod with the same SLA as our stock models. Typical onboarding: two weeks from weights delivery to production cut-over. Your model never leaves your pod — we don't inspect weights and we don't aggregate across tenants.
Tiers are priced at baseline concurrent programs. Burst capacity is billed per program-hour at a published overage rate (documented in the MSA, not buried in a credit system). Your AP team sees a monthly invoice with baseline + burst lines, not a per-minute per-destination labyrinth. If your event is large and predictable, we'll pre-reserve capacity at the quoted baseline rate — ask.
Not a salesperson. An engineer who has deployed Facecast into an MCR. They will answer your procurement team's questions and run a test cut against your actual signal chain. No credit-card wall. No aggressive follow-up.
BOOK A DEMO