Nova Canvas Launches Virtual Try-On

🔑 Enhanced Key Takeaways

• •Amazon Nova Canvas virtual try-on was first added in July 2025, enabling realistic overlay of products onto people or spaces by combining source, reference, and mask images.[3]
• •The feature uses image-guided inpainting tuned for garments, accessories, furniture, and generalizes to logos or text, supporting up to five generated images per API call without text prompts.[4]
• •It offers options to keep or regenerate poses, hands, or faces in the source image, plus three image stitching modes: BALANCED, SEAMLESS, and DETAILED.[4]

🛠️ Technical Deep Dive

• •Requires three inputs: source image (e.g., person or room), reference image (product like jacket or couch, can include multiple outfit items), and mask image for guidance.[4]
• •Supports automatic removal of pose/hands/face from mask when choosing to keep them; no text or negative prompts supported, unlike other Nova Canvas tasks.[4]
• •Image stitching modes include BALANCED (default balance), SEAMLESS (smooth blending), and DETAILED (preserves fine details with potential seams).[4]
• •Integrated in serverless AWS architectures using Bedrock, Lambda, Step Functions, S3, DynamoDB, and SQS for scalable ecommerce deployment.[5]

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline