Skip to content

Tech Stack

CreatorStudio is a Studio-layer product built on top of the best available AI models, a modern edge-deployed web runtime, and Cloudflare’s developer ecosystem. The stack is organized in four layers: the AI models orchestrated, the application runtime, the infrastructure, and the publishing integrations.

Fifteen-plus models are routed through Agent Ra. The creator never picks one directly. Ra selects per shot, based on visual quality, narrative continuity, cost, latency, and outcome data from the Director Memory Graph.

CategoryModelsRole in the pipeline
Video (keyframe-to-video, motion)Kling 2.5, Runway, Pika, Luma, Seedance (video), HailuoPer-shot motion generation from approved keyframes
Video (frontier, long context)Google Veo 3, OpenAI Sora 2High-fidelity shots, longer durations, complex motion
Image (keyframes and stills)FLUX 1.1, Seedance (image)Per-shot keyframe generation before motion
Dialogue and voiceElevenLabsPer-character voice cloning, dialogue, continuity across scenes
Audio (score, ambient, supporting)MINIMAX, supporting audio modelsScore, ambient beds, SFX
Additional supporting modelsUpscalers, captioning, miscStage-specific enrichment

The stack totals fifteen-plus models and is designed to grow. When a better model ships, Ra adds it. The creator does not see a model swap. Orchestration detail lives in Model Orchestration.

The application is a modern web stack with an edge-deployed runtime. Specific framework versions are not listed here to avoid fabrication; the shape is conservative and standard for a 2026 edge-first product.

LayerChoice
Front endTypeScript single-page app, component-driven
RuntimeEdge-deployed runtime for global low latency
Document storeManaged relational store for creator accounts, projects, renders, billing
Vector storeEmbeddings store for the Director Memory Graph (voice, visual, audience, outcome signals)
Object storeManaged object storage for media artifacts (keyframes, shots, audio, renders)
Job orchestrationQueue-backed workers for the six-stage render pipeline

For the render pipeline itself, see Video Pipeline.

CreatorStudio publishes on the Cloudflare ecosystem. Cloudflare gives the product edge-first delivery, low cost-per-request, and a tightly integrated set of primitives (Pages, Workers, R2, D1) that keep the orchestration tier light.

PrimitiveRole at CreatorStudio
Cloudflare PagesMarketing site and static front ends (including creatorstudio.media and documentation)
Cloudflare WorkersApplication logic, API endpoints, request routing, auth, orchestration glue
Cloudflare R2Media artifacts: keyframes, per-stage shot outputs, final renders, per-platform variants
Cloudflare D1Relational data: accounts, projects, renders, billing, publishing calendars

GPU-heavy model inference is not self-hosted. It is provider-side, across the fifteen-model stack. That is deliberate. CreatorStudio’s orchestration tier is light, so margin widens as model costs commoditize (roughly 30 to 50 percent every six months at the model layer). The full infrastructure rationale lives in Infrastructure.

CreatorStudio ships to every platform a storyteller actually uses. The Publishing module handles calendar scheduling and 24x7 stream automation on top of these integrations.

PlatformWhat the integration does
YouTubeLong-form uploads, metadata, thumbnails, 24x7 Live Studio streams
TikTokShort-form variants from the same render, per-platform aspect ratios
InstagramReels, Stories, feed variants, captions
XClips, vertical and square variants, autoposting
LinkedInLong-form and short-form uploads, B2B formats

Subtitle Studio plugs in before Publishing to handle auto-captions plus multilingual dubbing, so one story ships in every language the creator needs.

The Channel Analyzer and Director Brief are free. Paid tiers are $20 one-time, Creator at $50 per month, Pro at $100 per month, and Power at $500 per month. No bring-your-own-key. The Studio ships with the full fifteen-model stack behind one subscription. Public beta opens April 20, 2026.