Amazon Nova: Hands-on learning

Amazon Nova is a new family of foundation models announced at Reinvent 2024. It delivers low cost, fast performance, and domain-specific customization with Rag for integration and responsible AI safeguards.

Explore the Amazon Nova family: content understanding models Nova micro, Nova Lite, Nova Pro for text, image, and video understanding; creative models Nova canvas and Nova Reel for content generation.

Compare Nova family members and their context windows. Choose between Nova micro, Nova Lite, Nova Pro, Canvas, or Nova Reel for text, image, or video generation with language support.

Access Amazon Nova and other foundation models via a single Bedrock API. Securely tailor models to your organization data, enable Rag, and perform API-driven tasks.

Explore Amazon Nova Canvas for image generation using Bedrock, SageMaker AI Studio, and Jupyter notebooks in a public or self-paced AWS workshop to boost creativity and cut production time.

Discover how to set up SageMaker Domain workshop in your AWS account and understand charges. Learn to request Nova Canvas model access in Amazon Bedrock and start using the model.

Set up a SageMaker domain in the AWS console, launch Studio, create a Jupyter Lab space, clone the repo, and begin image generation with Noah Canvas and bedrock.

Run notebooks to generate images from prompts with Amazon Nova Canvas. Upload notebooks to the resources folder and create data and output folders.

Generate images with amazon bedrock using text to image, inpainting, and outpainting; explore variations and background removal from console or api for fashion design use cases.

Learn how inpainting edits images by supplying an input image and a mask to replace objects, manipulating areas with a mask image, prompts, and seeds to generate new visuals.

Master outpainting by preserving a region with a mask or natural-language prompt and replacing the background to create new scenes with Amazon Nova models.

learn to remove backgrounds from images using a dedicated background removal task, upload your image, run remove background, and obtain an isolated object ready for new content or composition.

Generate image variations from an input image by guiding style with prompts or reference images, and adjust similarity strength and cfg scale for wedding dresses or tuxedos.

Learn image conditioning to generate outputs that follow a reference image using canny edge or segmentation, guided by a text prompt and adjustable control strength.

Use color conditioning to control image palettes with exact hex codes for your brand. Generate images featuring paisley, striped, and floral patterns on a white background using your specified colors.

Explore Amazon Nova capabilities through hands-on image generation workflows and color-guided variations. Use notebooks and workshop examples to perform image transformations like background removal, inpainting, and object replacement.

Learn how to clean up AWS resources after the workshop by stopping and deleting SageMaker domains, studios, and associated resources to avoid costs, and review Bedrock and Nova usage.

Learn to generate studio-quality 4K short videos with Amazon Nova Reel using text or image prompts. Control camera moves through prompts and manage async delivery to an S3 bucket.

Learn to generate videos from text prompts using Amazon Nova Reel in a hands-on workshop, covering text to video, image to video, and S3 storage with asynchronous status tracking.

Produce videos from images using Amazon Nova Reel by inputting an image and a prompt. Learn to chain and stitch clips to extend videos and create dolly in camera motion.

Learn prompting best practices for image and video generation with Amazon Nova and Nova Reel. Craft prompts describing subject, lighting, style, and camera motion, and use negative prompts and seeds.

Amazon Nova pricing uses token counts to calculate input and output costs, estimate image and video charges via the Bedrock service, and set budgets using the Amazon Web Services calculator.

Explore Amazon Nova content understanding models—NOAA micro, NOAA Lite, and NOAA Pro—and their multimodal capabilities for summarization, question answering, image captioning, OCR, and video analysis.

Explore Amazon Bedrock Nova models with hands-on examples for images, videos, and documents, covering formats, size limits, S3 integration, and document extraction plus image description.

Select the Amazon Nova model by balancing input type and priorities such as cost, speed, and efficiency, using multiple models for text, image, and video tasks on AWS Bedrock.

Explore customization approaches for foundation models, balancing cost, time, quality, and complexity. Use prompt engineering, retrieval augmented generation, and fine tuning to ground results in enterprise data.

Explore prompt engineering techniques for understanding and reasoning models using system, user, and assistant roles to structure inputs. Apply prompts and output constraints to improve clarity and reduce token cost.

Explore prompt engineering techniques to steer model output, from natural language versus programmatic formats to using section prompts, delimiters, and precise constraints for classification and structured results.

Learn how to use system prompts to set context, persona, and guardrails, control output formats like yaml, and prevent leakage of confidential information in model responses.

Explore zero-shot prompting, one-shot prompting, and few-shot prompting with diverse examples to improve model output. Practice chain-of-thought prompting to guide step-by-step reasoning and ground answers to reference text.

Discover how foundation models are created from diverse data, curated and tokenized, then trained through pre training and post training, with teacher and student models and distillation.

Explore Amazon Nova training infrastructure, including Amazon EC2 accelerated instances, Elastic Fabric Adapter, S3 and FSx for Lustre storage, and Amazon SageMaker Hyper Pod, plus orchestration and monitoring.

Access Amazon Nova models on Bedrock via converse, invoke, and streaming APIs for stateful chats or single prompts, with no long-term data retention.

Set up a self-paced workshop to access Amazon Nova models via bedrock and SageMaker AI, using Jupyter Lab notebooks to invoke APIs and explore multimodal text and image understanding.

Learn to set up a Python virtual environment and install boto3, then invoke Amazon Nova multimodal models via Bedrock, handling images, video, and S3 inputs.

Explore repeatable patterns with Amazon Nova, using charts and images to extract data via Bedrock Nova models, store in S3, query with Athena, and explore multi-modal tools.

Explore Amazon Nova use cases, options, and best practices, then complete the workshop to reinforce knowledge while considering costs, and leverage resources and teaching others to deepen learning.