Fun A.I. Moments is a Twitch extension that lets viewers spend Bits to transform a livestream in real time using AI-generated visual effects. Viewers type a creative prompt, the system captures the live video feed, runs it through a StreamDiffusion pipeline on GPU, and broadcasts the transformed result back to all viewers as a shared "moment." The project integrates Daydream's Scope plugin for real-time video ingest, a custom bridge page for WebRTC streaming, and a full cloud-native backend on AWS (EKS, SQS, DynamoDB, S3).

Development Milestones (Feb 11 - Feb 21, 2026)

Feb 11 - Built the foundational SKU 2 on-demand moments system with Twitch Bits integration, the streamer bridge page for live camera capture, Scope plugin integration for WebRTC video ingest, and OAuth-based authentication for the Twitch API.

Feb 12 - Implemented dynamic GPU lifecycle management with RunPod serverless infrastructure, deployed the moments API and worker services to EKS, and established the WebRTC pipeline connecting the bridge page to the StreamDiffusion backend.

Feb 13 - Migrated to Daydream V2 bridge architecture using WHIP/WHEP protocols, integrated the SDXL Turbo pipeline for higher quality image generation, and tuned pipeline parameters for real-time performance at interactive frame rates.

Feb 15 - Launched V3 on-demand moments architecture with JWT-based session authentication, CORS security policies, viewer display name resolution from Twitch, and a complete rework of the moment submission flow for reliability.

Feb 16 - Added DynamoDB-backed durable moment queue for persistence across restarts, built a moments reaper service for automatic cleanup of stale entries, introduced the queue management UI (V3.4), and added pre-warm controls so streamers can keep the GPU pipeline hot before going live.

Feb 17 - Implemented bridge-driven playback where the streamer's bridge page controls moment lifecycle, built a dedicated moderator dashboard for queue management and moment approval, and resolved critical race conditions in the WebSocket event flow.

Feb 19 - Integrated Twitch chat notifications so viewers see real-time updates when moments are submitted and completed, and added WebSocket health check and automatic reconnection logic for the bridge page.

Feb 20 - Added a pipeline settings UI for streamers to tune StreamDiffusion parameters in real time, explored IP-Adapter and FaceID integration for identity-preserving transforms, implemented per-channel OAuth tokens for Twitch chat, and switched to the SDXL model for improved visual quality.

Feb 21 - Built the scene frame capture pipeline: the bridge page now periodically captures camera frames and uploads them to S3 via presigned URLs, enabling the worker to analyze scene content for future object detection and contextual effects. Added full Terraform CI/CD automation for the new S3 infrastructure.