light-heart-labs--dreamserver/ods/extensions/library/services/forge

Forge / A1111

The original Stable Diffusion web UI (based on Automatic1111). Features extensive model support, extensions, inpainting, outpainting, and professional-grade image generation capabilities.

Hardware compatibility

This extension ships a pinned image digest. The bundled Torch + CUDA build inside that image is not portable across every NVIDIA generation, so compatibility is constrained by the image rather than ODS itself.

• NVIDIA-only. The manifest declares gpu_backends: [nvidia], so this extension is filtered out at install time on AMD, Intel, Apple Silicon, and CPU-only systems. Apple Silicon and CPU-only hosts cannot install it.
• Confirmed broken: NVIDIA RTX 3070 Mobile (compute capability sm_86) under CUDA driver capability 12.4. The container starts and the host shows it as running, but the Forge process inside crashes with Your device does not support the current version of Torch/CUDA. The pinned Torch wheel in the bundled ghcr.io/ai-dock/stable-diffusion-webui-forge image is incompatible with this hardware/driver combination.
• Untested: all other NVIDIA hardware. RTX 30-series desktop parts, other sm_86 mobile parts, RTX 40-series (sm_89), Hopper (sm_90), and earlier generations have not been live-tested with this image. We make no prediction about whether they work or fail - treat them as unknown until tested on real hardware.
• Linux: the live-confirmed failure above was reproduced on Linux with the NVIDIA Container Toolkit. Non-sm_86 Linux hosts are untested.
• Windows / WSL2: the same image runs under WSL2 with nvidia-container-toolkit. Because the Torch/CUDA mismatch is internal to the image, a WSL2 host on the same RTX 3070 Mobile (sm_86) silicon is inferred to hit the same crash - but this has not been live-tested. Treat as inferred, not confirmed.
• macOS: Apple Silicon and Intel Macs cannot run this extension. The manifest's gpu_backends: [nvidia] filter removes it from the install set on macOS.

If you hit the incompatibility on supported-but-unlisted hardware, the upstream project is lllyasviel/stable-diffusion-webui-forge; building a compatible image yourself or substituting an alternative Forge image is currently the only path forward, and is outside the scope of this packaged extension.

Privacy & Defense-in-depth

The upstream ai-dock Forge image bundles syncthing, quicktunnel, serviceportal, and sshd for cloud workflows, plus a jupyter notebook server. These are autostarted by the image's bundled supervisord and would otherwise expose the host to remote relays, tunnel endpoints, inbound SSH, and an unauthenticated notebook on default install.

ODS disables them via SUPERVISOR_NO_AUTOSTART=syncthing,quicktunnel,serviceportal,sshd,jupyter in compose.yaml. To re-enable any of them, override that env var in your .env or a compose override.

cloudflared is also bundled but only autostarts when CF_TUNNEL_TOKEN is set, so it stays a no-op by default.

Requirements

• GPU: NVIDIA (min 8 GB VRAM)
• Dependencies: None

Enable / Disable

ods enable forge
ods disable forge

Your data is preserved when disabling. To re-enable later: ods enable forge

Access

• URL: http://localhost:7861

First-Time Setup

1. Enable the service: ods enable forge
2. Open http://localhost:7861
3. Start generating images with the txt2img tab

First startup may download several GB of model files. Subsequent starts are instant.

API Usage

# Generate via txt2img API
curl -X POST http://localhost:7861/sdapi/v1/txt2img \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "a photo of a cat in a garden",
    "steps": 20,
    "width": 512,
    "height": 512
  }'

# Check progress
curl http://localhost:7861/sdapi/v1/progress