In progress
3Sora 2 direct integration
Native Sora 2 support without the fal.ai relay — lower latency, direct status polling, and full prompt conditioning. Waiting on OpenAI partner API access.
Social scheduling (Postiz)
Publish rendered videos directly to TikTok, Instagram, and YouTube from the Brand Studio. Powered by a dedicated Postiz instance with cron-driven post queues.
Operator dashboard
Multi-tenant control plane for agencies and platforms. Manage sub-accounts, delegate brands, inspect job traces, and set per-account spend caps from a single admin view.
Planned
4Runway and Google direct routes
First-party integrations with Runway Gen-4 and Google Veo 3 — no intermediary relay, direct API keys, full model parity.
iOS push notifications (Operator tier)
Native push alerts when long-running renders complete. Operator-tier feature for production teams running overnight batch jobs.
Agent memory & project context
Persistent per-project memory that agents read before starting a run — brand voice, past performance notes, blacklisted assets. No more re-briefing on every session.
Real-time render streaming
Stream partial frames to the client over SSE as the model generates — no more polling loops. Starts with Seedance and Kling where the provider supports progressive output.
Shipped
8MCP server — 35 tools
Full Model Context Protocol endpoint at /api/mcp. Claude Desktop, Continue.dev, and any MCP-compatible agent can generate video, images, voice, and music with a single bearer token.
BYOK + platform credits
Bring your own provider API keys (Kling, fal, ElevenLabs, Google, OpenAI) at list price with zero markup, or use Varosity Credits and let us handle the key management.
Avatar lip-sync (OmniHuman)
Drop a photo and a voiceover clip; get back a talking-head video. Full-replacement and picture-in-picture modes. Used in Brand Story and Operator flows.
Brand agents + creative bible
Per-brand style bible, asset library, and an autonomous agent that runs end-to-end creative runs without human approval gates.
Storyboard mode (v3 schema)
Shot-level style, asset, and frame-chain composition. Agents plan, generate grid thumbnails, and render the full storyboard in one skill call.
Self-updating agent skills
Skill docs live at /api/v1/skills/ and carry an auto_update flag. Agents call refresh_skills at the start of each session to pull the latest workflow spec without a redeploy.
Outbound webhooks with retry queue
Register HTTPS endpoints to receive shot.generation.completed, render.completed, voice.completed, and music.completed events. Exponential-backoff retry (up to 5 attempts) backed by a persistent queue.
Subscription tiers (Solo / Studio / Operator)
Monthly plans with per-brand caps, priority routing, and BYOK-required gates. Stripe Checkout handles payment; credits are granted atomically on checkout.session.completed.