Iris
Launch Studio
Real-time multimodal creative studio

See. Speak. Create.

Iris is a voice + vision AI studio that watches what you see, understands what you say, and generates cinematic visuals in real time — with production-grade latency and cost observability.

Voice

Streaming ASR + TTS

Vision

Camera, screen, upload

Creation

Generate & edit images

Capabilities

Built like a real product, not a tutorial

Voice-first interaction

Push-to-talk with Whisper transcription and natural TTS narration of results.

Live vision input

Camera, screen share, or image upload — Iris sees what you see.

Streaming pipeline

SSE events stream thinking states, tool calls, and visuals with sub-second perceived latency.

Tool-calling LLM

GPT-4o decides when to generate, edit, or describe — no prompt hacking.

Production observability

Track tokens, cost per session, p50/p95 latency, and GPU seconds in real time.

Shareable replays

Save and share creative sessions with public replay links.

Architecture

Real-time multimodal pipeline

  1. 1User speaks + shares vision frame
  2. 2Whisper ASR transcribes audio
  3. 3GPT-4o orchestrator routes via tool calling
  4. 4SDXL / DALL-E generates or edits visuals
  5. 5TTS narrates results back to user