"Put a tiny cowboy hat on the cat"
|
"Make the cat wear a formal black tuxedo with a bow tie"
|
"Add cool aviator sunglasses"
|
"Add a red superhero cape"
|
"Put a royal golden crown on the cat"
|
"Turn this into an oil painting in renaissance style"
|
"Change background to outer space"
|
"Put in a cozy living room with fireplace"
|
"Change background to snowy winter wonderland"
|
> All 9 edits generated via the production API in ~3.5s each on H100. Full API docs in [`firered-lora/`](firered-lora/).
---
## Architecture
```
┌─────────────────────────────┐
│ maya-baseten-deploys │
│ (this repo) │
└──────┬──────────┬────────────┘
│ │
┌──────▼──┐ ┌────▼─────┐
│firered- │ │ veena/ │
│ lora/ │ │ (next) │
└────┬────┘ └──────────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
┌────────┐ ┌────────┐ ┌────────┐
│ model/ │ │ api/ │ │samples/│
│ Truss │ │FastAPI │ │ demo │
│(Baseten│ │(Vercel)│ │ images │
│ GPU) │ │ │ │ │
└────────┘ └────────┘ └────────┘
```
## How it works
1. **Each deployment** lives in its own folder with a Truss model (`model/`) and API server (`api/`)
2. **Push to Baseten** — `truss push` deploys the GPU model
3. **Push to Vercel** — `vercel --prod` deploys the API layer
4. **Scale from Claude Code** — hit `/api/gpu/scale?replicas=N` to add/remove GPUs
## Adding a new deployment
```
mkdir new-model/
├── model/
│ ├── __init__.py
│ └── model.py # load() + predict()
├── config.yaml # Baseten Truss config
├── api/
│ ├── server.py # FastAPI endpoints
│ └── requirements.txt
├── samples/ # Demo images
└── README.md
```
---
Built by [MayaResearch](https://github.com/MayaResearch)
