GNSS — blockchain meets generative AI

MEMs recorded in the project snapshot: 4,339
Reported generation time during the project: 6–12s
public installation: Lisbon

GNSS is a generative-art ecosystem where Ethereum ERC-721 species tokens provide the traits used to generate new Machine Embedded Memories (MEMs). I built the applied AI system around a fine-tuned Stable Diffusion model, from model experimentation and inference serving to the generation interface and lineage explorer. MGXS provided the collection’s visual identity and artistic direction.

Generation and lineage

Traits supplied conditioning vocabulary for MEM generation. Seed and prompt construction related each output to species metadata; lineage records connected memories to ancestors in the Tree of MEMs.

Serving inference

Python and Jupyter supported research and evaluation. Model artifacts and a custom inference script were packaged in S3 for SageMaker deployment. API Gateway and Lambda handled requests to the GPU endpoint, separating model deployment from request serving. Reported generation time during the project was 6–12 seconds; the interface replayed the previous MEM during the initial wait.

AWS architecture

CloudFront, API Gateway and Lambda connect the experience to SageMaker. S3 and CloudFront serve assets and imagery; MongoDB stores species metadata; Firebase AppCheck gates client calls; Secrets Manager manages credentials; GitHub supports delivery. Cloudflare provides DNS. The architecture views separate request serving, model deployment and metadata relationships.

The client reaches a serverless API through CloudFront. Lambda shapes the request and invokes SageMaker; metadata and generated output use separate stores.

01
MEM client CloudFront
request
02
CloudFront API Gateway
API request
03
API Gateway AWS Lambda
invoke handler
04
MongoDB AWS Lambda
trait context
05
AWS Lambda SageMaker
inference input
06
SageMaker AWS Lambda
generated output

MEM client · Web experience / installation

Audience and collection interactions enter through the web experience. Cloudflare provides DNS.

CloudFront · Edge delivery

CloudFront serves the experience and fronts the request path.

API Gateway · REST entry point

The API is the public application boundary in front of Lambda.

MongoDB · Species metadata

Collection traits and lineage context are stored separately from GPU inference.

AWS Lambda · Request shaping

Application logic prepares model input and invokes inference. Firebase AppCheck gates client calls; Secrets Manager manages credentials.

SageMaker · Stable Diffusion endpoint

Custom inference code and model artifacts serve image generation.

S3 + CloudFront · Generated art / web assets

Storage and delivery serve generated output and the ecosystem’s front ends.

Design decision

Place GPU inference behind the application API. Cloudflare handles DNS, while AppCheck and Secrets Manager support access and credentials.

Conceptual architecture. Explore component details in the diagram or the connection list.

NFC Lisbon ’23

At the Lisbon installation, attendees supplied input and watched generated species emerge.

Collection

The collection uses Ethereum ERC-721 tokens whose metadata points to the ecosystem API. MongoDB-backed traits and S3-hosted art support the collection; OpenSea and Rarible consume them for listings. The project snapshot records 4,339 MEMs.