Developer API

Deploy visual pipelines with our API Engine.

A predictable layout and rendering infrastructure with unified SDK client packages, webhooks, and structured response streaming. Built to handle production traffic.

Get API key Read docs

curl https://api.veltneon.com/v1/images \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "a futuristic city at golden hour, cinematic",
    "style": "photoreal",
    "resolution": "4K"
  }'

200 OK · Render Succeeded

response_trace.json

{
  "id": "img_7x8932pq19",
  "status": "succeeded",
  "created_at": 1782439120,
  "completed_at": 1782439121,
  "model": "veltneon-lumen-v2",
  "style": "photoreal",
  "resolution": "4096x2160",
  "output": {
    "url": "https://cdn.veltneon.com/renders/7x8932pq19.png",
    "mime_type": "image/png",
    "size_bytes": 4821034
  },
  "inference_metrics": {
    "compute_node": "NV-H100-DGX-04B",
    "engine": "TensorRT / Triton",
    "latency_ms": 1940,
    "precision": "FP8",
    "cost_tokens": 12
  }
}

SDKs

Official SDK libraries.

JavaScript / TypeScript

npm install @veltneon/sdk

Python

pip install veltneon

Go

go get github.com/veltneon/go

Event Hooks

Webhook lifecycle flow.

Integrate asynchronous callbacks to resolve high-resolution renderings inside your database records.

1. API Request

Payload includes `webhook_url` endpoint

2. Render Queued

Gateway posts `render.pending` check

3. Processing

H100 Node resolves latent canvas layers

4. Success Dispatch

Gateway posts `render.succeeded` with PNG URL

SSE Stream

Server-Sent Events (SSE)

For real-time UI previews, developers can listen to layout token changes progressively using our EventStream API path. Progressive previews are yielded while denoising passes execute.

event_stream_terminal

// Click simulate to start listening...

composition_layers.json

{
  "layers": [
    {
      "id": "bg_layer_01",
      "type": "background",
      "prompt": "studio soft ambient background",
      "weight": 0.85
    },
    {
      "id": "product_mesh_01",
      "type": "foreground_silhouette",
      "source_contour_url": "https://assets.brand.com/bottle_mesh.obj",
      "position": { "x": 0.5, "y": 0.6, "scale": 1.2 },
      "light_angle": 120
    }
  ],
  "global_overrides": {
    "color_palette": ["#8B5CF6", "#EC4899"],
    "locked_contours": true
  }
}

Payload Schema

Dynamic Composition Objects

veltneon supports rich layer lists where objects hold dynamic bounds, reference shapes, and custom light angles. This structured JSON representation parses directly to intermediate tensor maps in Triton nodes.

Rate Limits

Token-bucket rate limits

In order to protect H100 computing channels, API keys are governed by Token-Bucket boundaries. Default sandbox keys process up to 10 image requests per minute with a bucket buffer size of 20 tokens.

X-RateLimit-Limit: 10/min

X-RateLimit-Remaining: 8

X-RateLimit-Reset: 14

GitOps Pipeline

CI/CD asset auto-compiles

Link visual updates directly into code reviews. Git hooks trigger veltneon automated API builds to test design layouts, check compliance boundaries, and render production assets automatically.

Offline Testing

Local Sandbox Emulator

Test your API routes without incurring server cost or hitting GPU queues. Our sandbox emulator package mocks all endpoints locally, returning mock schemas and layout contours.

npx @veltneon/emulator start

Endpoints

Simple API endpoints.

POST/v1/imagesGenerate a layout from composition specifications.

POST/v1/editsApply layer edits, inpainting, or scene expansions.

POST/v1/variationsRender design style and lighting variations.

GET/v1/jobs/:idRetrieve batch status for asynchronous pipelines.

POST/v1/finetunesRegister a style palette fine-tune configuration.

NVIDIA Hopper Architecture

Compute Specs

Cluster HostDGX-H100-Node-04

Compute CoreH100 Tensor Core

Precision EngineFP8 Mixed Precision

Inference CompilerNVIDIA TensorRT

Active Uptime99.998% SLA

Hardware Acceleration

NVIDIA H100-DGX Accelerated Backend.

We compile and scale generative models directly on enterprise-grade NVIDIA DGX compute nodes. By leveraging CUDA-X deep learning layers, custom FP8 precision models, and NVIDIA Triton Inference Server, veltneon renders multi-layer 4K image assets in less than 2 seconds.

TensorRT CompiledTriton Dynamic BatchingCUDA Kernels Optimized

Execution Pipeline