Run Your Higgsfield Code on LivePair — a Compatible Endpoint
LivePair AI ·
Higgsfield opened its developer surface this year — an API, SDKs, a CLI, agent skills — and a small ecosystem of third-party tools already speaks its protocol. LivePair now answers that same protocol at a drop-in endpoint: point a Higgsfield client at https://livepairai.com/hf and it runs on LivePair's model catalog, billed per result to a prepaid key. This is a compatibility layer, not an affiliation — it exists because the wire format is the ecosystem's real interface. Disclosure: LivePair is our platform.
What 'compatible' means here
Two Higgsfield wire formats exist in the wild, and /hf speaks both. The platform API used by community studios like Open-Higgsfield submits to POST /{model-path} and polls GET /requests/{id}/status, authenticated as Authorization: Key id:secret. The official SDK family (higgsfield-js, higgsfield-client, the CLI) posts to /v1/{task}/{model} and polls /v1/job-sets/{id} with hf-api-key and hf-secret headers.
A request to either shape gets translated and dispatched to LivePair's native generation path — the same billing, policy checks and metering as /v1/agent. Your prepaid lp_ key goes in either credential slot: Key lp:YOUR_KEY, hf-secret: lp_…, or a plain Bearer token all resolve.
Running Open-Higgsfield on LivePair
The community-built studio (one prompt bar, per-model settings rails, a shared gallery) is BYOK against a configurable API origin — the cleanest possible integration surface. Two changes run the whole app on LivePair:
| Setting | Higgsfield default | LivePair |
|---|---|---|
| HF_API_BASE_URL | the platform origin | https://livepairai.com/hf |
| Platform key | id:secret from Higgsfield | key:lp_… (either slot works) |
What maps where
Model paths translate by surface rather than by name — Kling and Seedance paths route to Wan 3.0 (image-to-video when a source frame rides along, text-to-video otherwise), image paths with an input image go to Seedream edit, Ideogram keeps its name, and the catch-all image default is Seedream 5.0. LivePair's own model ids also pass straight through, so POST /hf/qwen-image-3 works for callers that already know the catalog.
Parameters follow the same rule: prompt, image_url/input_images, aspect_ratio, duration, resolution and audio flags carry over; higgsfield-only knobs — elements, soul ids, motion presets, batch_size — are dropped instead of faked.
Honest limits
Compatibility ends where Higgsfield's product surface begins. Soul ID character training, the elements reference system, motion-control presets, custom references, speak/voice tools and webhooks have no LivePair equivalent and return a clear 501 rather than a silent mismatch. What you get on LivePair's side is its own lineup — a private model tier, per-result billing with no subscription, x402 wallet payment without an account, and renders that auto-delete after 48 hours.
If your workflow depends on a specific Kling or Soul render, stay on the real thing. If what you need is the client's workflow — a prompt bar, a settings rail, a gallery, an SDK call — with different economics and a different model line, the endpoint is live.
Why ship a compat layer
Because the ecosystem's interface is the wire format, not the brand. OpenAI-compatible, S3-compatible, and now Higgsfield-compatible are the same move: meet developers where their existing code already points. The full native surface — catalog, jobs, tools, OpenAI image/video shapes, MCP — stays at /v1/agent; /hf just removes the rewrite for anyone arriving with a Higgsfield-shaped client.