Status: [UNTESTED]

Endpoints found:
http://localhost:4000/v1/images/generations
http://localhost:4000/v1/videos/generations
http://localhost:4000/v1/models
http://localhost:4000/v1/stats
http://localhost:4000/v1/logs?page=1&per_page=50

Source: https://github.com/visgotti/litegen

📖 **Docs, SDKs & API reference → [visgotti.github.io/litegen](https://visgotti.github.io/litegen/)**

# LiteGen

> The universal proxy for AI image and video generation. Like LiteLLM, but for multimedia.

LiteGen provides a unified API gateway for all major image and video generation providers. Route requests across OpenAI DALL-E, Stability AI, Replicate, Google Imagen, Fal.ai, Runway, Luma, and more — with automatic fallback, weighted load balancing, caching, cost tracking, and a real-time dashboard.

## Features

- **Unified API** — OpenAI-compatible REST endpoints for image & video generation
- **10+ Providers** — OpenAI, Stability, Replicate, Google, Fal, Runway, Luma (more coming)
- **Smart Routing** — Fallback chains, weighted round-robin, lowest-cost, lowest-latency
- **Caching** — In-memory cache with configurable TTL to avoid duplicate generations
- **Cost Tracking** — Per-request cost estimation and aggregate spend analytics
- **API Key Management** — Create/revoke keys, weighted key pools for rate limit distribution
- **Observability** — Request logging, Prometheus metrics, structured tracing
- **React Dashboard** — Real-time monitoring, provider health, request logs, cost charts
- **Easy Config** — YAML config + env vars, auto-discovers provider keys
- **Self-Hosted** — Single binary + Docker, SQLite or Postgres

## What's new in v0.2

- **Capability registry with per-model schemas** — YAML-driven model declarations define allowed parameters, sizes, aspect ratios, and reference-image roles; unknown params are rejected (strict mode) or silently dropped (lax mode).
- **`reference_images` tagged-union** — Send reference images as URLs, base64 blobs, or multipart file uploads using `{type: "url"|"base64"|"blob", value: "..."}`.
- **Strict validation with per-request lax opt-out** — Default `strict: true` rejects unsupported parameters; set `strict: false` to drop unknowns and get an `X-Litegen-Dropped-Params` response header.