# Guidance for an AI-Powered Studio on AWS ## Overview This Guidance helps studios create cinematic-level AI-generated content at scale while maintaining full IP provenance and creative consistency across all assets. A unified artist interface abstracts complex AI tools into a streamlined experience, automatically tracking asset versions and lineage as creative teams work. The platform supports both hosted models through Amazon Bedrock and Amazon SageMaker AI, as well as third-party models, with all configurations managed centrally so artists can focus on creativity rather than tooling. You can produce high-quality creative content faster with confidence that every asset is traceable, governed, and built on a scalable foundation that grows with your studio's needs. ## Benefits ### Unify AI tools for artists Consolidate multiple AI content generation models into a single governed interface. Enable your creative teams to produce images, video, and audio without managing disparate tools or infrastructure. ### Track content lineage automatically Capture full provenance for every AI-generated asset, including models, parameters, and inputs used. Reduce IP and compliance risk with immutable audit trails stored in Amazon DynamoDB. ### Scale GPU inference on demand Leverage AWS Deadline Cloud to provision GPU infrastructure only when needed for parallel inference workloads. Avoid fixed compute costs while supporting studio-wide content generation at scale. ## How it works ### Platform overview These technical details feature an architecture diagram to illustrate how to effectively build this solution. The architecture diagram shows the key components and their interactions, providing an overview of the architecture's structure and functionality step-by-step. [Download the architecture diagram](downloads/ai-powered-studio-on-aws.pdf)Step 1Amazon CloudFront distributes the web application globally for scalable low-latency access to users via web browser. AWS IAM Identity Center authenticates and authorizes users through unified single sign-on (SSO).Step 2The Asset Management Service tracks all versions of every asset added or generated through the platform. It provides full audit trails and carries lineage, captured automatically as artists work (see Asset Management breakout diagram).Step 3The Artist Interface abstracts AI tools into a unified experience bounded by governance controls.Step 4The Content Generation Service (see Content Generation diagram) processes and routes content creation requests to the appropriate model endpoints for execution.Step 5All assets are stored in Amazon Simple Storage Service (Amazon S3), with unique identifiers to track versions and lineage. Amazon DynamoDB stores lineage and records correlated with each asset.Step 6AWS AppConfig stores configurations on AI models, allowing users to select their model of choice. This allows proper API calls and prompts to be used with both hosted and 3rd Party models, transparently to the end-user.Step 7Amazon SageMaker AI fine tunes custom and open-sources models using Low-Rank Adaptation (LoRA) training. It stores trained models and LoRAs securely in Amazon S3.Step 8Invoke third-party models via API. Store returned content in Amazon S3 through the Asset Management Service.Step 9AWS Deadline Cloud loads the LoRAs locally and scales on managed GPU infrastructure for large-scale inference generation.### Artist interface service architecture These technical details feature an architecture diagram to illustrate how to effectively build this solution. The architecture diagram shows the key components and their interactions, providing an overview of the architecture's structure and functionality step-by-step. [Download the architecture diagram](downloads/ai-powered-studio-on-aws.pdf)Step 1Custom UX can provide artists tools to meet customer specific needs. Examples include node-based workflow editors, configuration and prompt management, script and asset review, AI playgrounds, and more.Step 2REST APIs present an interface for the UX to execute individual functions.Step 3AWS Lambda functions execute small, single-purpose operations such as calls to content generation APIs.Step 4WebSocket connections push job progress and results back to users in near real-time, eliminating the need to poll for updates.Step 5Amazon DynamoDB provides single-digit-millisecond NoSQL storage for the control plane. Tables hold LoRA instance references, active WebSocket connections, and access controls.Step 6Amazon EventBridge emits an event upon render completion, delivering results back to users the moment a job finishes.### Asset management service architecture These technical details feature an architecture diagram to illustrate how to effectively build this solution. The architecture diagram shows the key components and their interactions, providing an overview of the architecture's structure and functionality step-by-step. [Download the architecture diagram](downloads/ai-powered-studio-on-aws.pdf)Step 1The Artist Interface calls Asset Management APIs to retrieve assets and store changes or updates.Step 2AWS Lambda functions retrieve each asset from Amazon S3 and make them available through Amazon CloudFront.Step 3Amazon S3 securely stores all assets. Each asset, including original content and any modified or generated version, remains available throughout production.Step 4Amazon DynamoDB stores provenance and the immutable audit trail. Every asset carries lineage (models, parameters, and inputs) captured automatically.Step 5Generated content (such as images, video, and audio) lands in an Amazon S3 bucket and is ingested by the asset management service. Amazon Simple Queue Service (Amazon SQS) handles event management and queuing. The ingestion process passes files to the primary asset bucket in Amazon S3 and writes metadata to Amazon DynamoDB.Step 6Amazon CloudFront serves multimedia assets from a private Amazon S3 bucket, providing fast delivery while restricting direct access to content.### Content generation service architecture These technical details feature an architecture diagram to illustrate how to effectively build this solution. The architecture diagram shows the key components and their interactions, providing an overview of the architecture's structure and functionality step-by-step. [Download the architecture diagram](downloads/ai-powered-studio-on-aws.pdf)Step 1AWS Lambda functions dispatch content generation processes initiated by artists.Step 2Amazon Bedrock provides access to large language models (LLMs) for textual analysis and script breakdown.Step 3Amazon DynamoDB stores job status and data while jobs execute. Copy completed job data and the generated asset are copied to the Asset Management Service.Step 4Amazon SageMaker AI fine-tunes hosted models such as Qwen and stores resulting LoRAs in a dedicated Amazon S3 bucket.Step 5AWS Deadline Cloud scales a GPU-based farm, allowing hosted models to run inference at massive parallel scale.Step 6Invoke third-party models such as Kling or Luma directly via API, or through model aggregators such as fal.ai, providing users access to required tools.Step 7Third-party models store generated images and videos in an Amazon S3 bucket. The Asset Management Service ingests those assets along with associated metadata.Step 8AWS AppConfig stores integration settings for self-hosted and third-party models as configuration. Users select models per task and administrators approve or block models through configuration settings.Step 9Agents monitor and evaluate provenance collection to capture all relevant lineage. They also retrieve any missing data and log it correctly.## Related content - **Guidance for an AI-Powered Studio on AWS**: This whitepaper provides detailed guidance on building an AI-powered studio platform on AWS for cinematic-level content generation at scale. [Download the whitepaper](downloads/ai-powered-studio-whitepaper.pdf) [Read usage guidelines](/solutions/guidance-disclaimers/)