Settings

SDXL · your machine

The AI engine is SDXL diffusion running on your own hardware via your sovereign API (port 8003). Three modes, one active at a time. No cloud providers, no third-party keys — the only secret here is the optional auth value for your own API, which stays on this server and is never sent back to any browser.

Engine modes

Live status, probed from the server just now. checking…

SDXL sovereign API priority

checking…

ComfyUI

checking…

Standalone fallback

checking…

Active mode

SDXL sovereign API ComfyUI Standalone

The worker only attempts the active mode. Queued jobs retry it with backoff while it's offline or warming up — nothing is ever submitted into a 503.

Sovereign API adapter

How the backend talks to your SDXL API: endpoint paths, HTTP method, optional header auth, and how our fields map to yours. Results stream back over SSE — the terminal event carries the image inline, so no second fetch is ever needed.

Adapter pre-filled from Matthew's SDXL API contract — edit only if his setup changes (e.g. the port moves).

Base URL

Your API as seen through the tunnel — usually the tunneled port on 127.0.0.1.

Health path

200 = ready · 503 = model still loading

Generate path

SSE stream path (use {job_id})

Best result path — the terminal event carries the image as base64.

Status path (fallback)

Result path (fallback)

Polled only if the SSE stream stalls.

Method

Auth header (optional — his API has none)

Auth value (optional, secret)

No auth value stored.

Field map (JSON object: our field → your field)

Our fields: prompt, negative_prompt, steps, guidance, width, height, seed, return_base64. Prompts are always truncated to 1000 chars — your API's hard cap.

Browser-direct checking…

Probing http://localhost:8003 from this browser…

If the tunnel is down but you're at your machine, the Create page can call your SDXL API straight from this browser (loopback — still your hardware). Your API needs CORS_ORIGINS to include https://psykomedia.com for this to work — see the setup guide.

How to connect

Full guide: LOCAL-FLUX-SETUP.md — SDXL API first, then ComfyUI, then the standalone fallback. Tunnel, GPU guidance, rate limits, troubleshooting.

Tunnel for the SDXL API (port 8003 on your machine):

ssh -N -R 127.0.0.1:8003:127.0.0.1:8003 root@209.46.125.134

Offline is fine. If your machine is off, the tunnel drops, or the model is still loading, SDXL jobs queue on the server and retry automatically with backoff — nothing fails silently, nothing gets faked. Meanwhile the procedural engine in Create works regardless.