Image & Vision AGPL-3

Stable Diffusion WebUI Forge

High-performance fork of Automatic1111 with aggressive low-VRAM optimizations. Best Stable Diffusion UI for 6-8GB cards running SDXL.

Website
Platforms: windowsmacoslinux

Stable Diffusion WebUI Forge is lllyasviel’s fork of Automatic1111, rewritten where it matters: the attention and memory-management backend. The UI looks almost identical to A1111 — same Gradio tabs, same prompt syntax, same extension hook points — so existing tutorials and muscle memory transfer directly. The difference is under the hood. Forge reworked how model weights are paged between CPU and GPU RAM, which means SDXL runs comfortably on 6GB cards that previously choked, and batch sizes on 8-12GB cards roughly double. It also added first-class support for Flux, SD3.5, and Hunyuan without the convoluted install guides those models required on vanilla A1111. If you want the A1111 experience on modest hardware or with newer architectures, Forge is the default recommendation in 2026.

Key Features

  • Drastically lower VRAM usage. SDXL generates 1024x1024 in ~5GB; Flux-schnell runs in 8GB. Forge’s weight-paging strategy keeps the GPU fed without the constant CPU<->GPU swapping that made A1111 slow on small cards.
  • Same UI and extensions as A1111. Most A1111 extensions install and run unchanged. Prompt weighting, XYZ Plot, Dynamic Thresholding, and the standard LoRA/embedding workflow all work.
  • Native Flux, SD3.5, Hunyuan support. No separate ComfyUI graph or manual clip/t5xxl juggling — pick the model in the checkpoint dropdown and generate.
  • Faster sampling on equivalent hardware. Benchmarks typically show 1.5-3x more it/s than A1111 on the same GPU, especially above 512x512.

When to Use Stable Diffusion WebUI Forge

Forge is the right pick when you want the classic A1111 interface but your GPU has 6-8GB of VRAM, or when you want to run Flux/SDXL without learning ComfyUI’s node graph. It is also the smoothest migration path for existing A1111 users: point Forge at your existing models/ folder and your old workflows keep working, faster. Choose ComfyUI instead if you need video, complex ControlNet routing, or shareable graph files; choose SwarmUI if you want multi-user access out of the box.

Setup

git clone https://github.com/lllyasviel/stable-diffusion-webui-forge
cd stable-diffusion-webui-forge
# Windows: run webui-user.bat
# Linux/macOS:
./webui.sh

# Point at an existing A1111 models folder instead of re-downloading:
# in webui-user.sh:
#   export COMMANDLINE_ARGS="--ckpt-dir /path/to/a1111/models/Stable-diffusion \
#       --lora-dir /path/to/a1111/models/Lora --embeddings-dir /path/to/a1111/embeddings"

Python 3.10 is recommended. The first launch downloads torch and builds the venv. For Flux, drop the .safetensors checkpoint into models/Stable-diffusion and select it — Forge handles the text encoders automatically.

See also