Key Takeaways
- Forge (github.com/lllyasviel/stable-diffusion-webui-forge) is free and open-source under the AGPL-3.0 license
- Built by lllyasviel, also known for creating ControlNet, as a fork on top of AUTOMATIC1111's Stable Diffusion WebUI
- Its README describes Forge as "a platform on top of Stable Diffusion WebUI" for resource management and inference speed, named after Minecraft Forge
- Rebuilds the WebUI's UNet-patching backend, targeting lower VRAM use and faster inference on the same hardware
- Ships built-in ControlNet and IP-Adapter support, LoRA support, and Flux model support (NF4/GGUF quantization) with no extra extensions needed
- Uses a Gradio 4 interface, with a Windows one-click package plus manual git-clone setup for macOS and Linux
- Runs on Windows, macOS, and Linux; a GPU is strongly recommended, with no single documented VRAM minimum β it depends on the checkpoint and resolution
π In One Sentence
Stable Diffusion WebUI Forge is a free, open-source (AGPL-3.0) fork of AUTOMATIC1111's Stable Diffusion WebUI, built by ControlNet creator lllyasviel, that re-architects GPU memory management for faster inference and lower VRAM use while keeping most of the original WebUI's extensions working.
π¬ In Plain Terms
Forge is the same basic Stable Diffusion image-generation web app you may know from the AUTOMATIC1111 WebUI, rebuilt under the hood so it runs faster and needs less graphics-card memory for the same image. You install it the same way β a one-click Windows package or a manual clone-and-script setup β and most extensions and checkpoints that work in the original WebUI work in Forge too.
πNote: This review is the deep-dive companion to Stable Diffusion WebUI Forge's entry in the Local LLM Software Directory β see that page for how Forge compares at a glance to dozens of other local AI tools.
What Is Stable Diffusion WebUI Forge?
Stable Diffusion WebUI Forge is a free, open-source local web app for generating images with Stable Diffusion and Flux models, built as a fork on top of AUTOMATIC1111's Stable Diffusion WebUI. Its own GitHub README describes it as "a platform on top of Stable Diffusion WebUI (based on Gradio) to make development easier, optimize resource management, [and] speed up inference," explicitly naming Minecraft Forge as the inspiration for the project's name and role.
- Core function: a browser-based UI, served locally from your own machine, for text-to-image and image-to-image generation with Stable Diffusion and Flux checkpoints
- Creator: lllyasviel, also the creator of ControlNet, one of the most widely used Stable Diffusion conditioning tools
- Foundation: built directly on top of AUTOMATIC1111's Stable Diffusion WebUI β Forge reuses that project's checkpoint format, much of its extension API, and its general UI layout, rather than rewriting the app from scratch
- Interface framework: Gradio 4, per the project's README and its listed known issues around a planned Gradio 5 migration
- License: AGPL-3.0, per the repository's LICENSE.txt file β a copyleft license, unlike the more permissive MIT/Apache licenses some other local AI tools use
- Canonical repository: github.com/lllyasviel/stable-diffusion-webui-forge, created January 14, 2024
Forge's Project History and Version Milestones
Forge's GitHub repository was created on January 14, 2024, and lllyasviel has continued pushing updates since, though at a visibly slower pace than in its first year. As of this review, GitHub's API shows the repository's most recent code push was July 31, 2025 β worth knowing if you are evaluating how actively maintained the project currently is.
- 1January 14, 2024 β Repository created
Why it matters: lllyasviel opens the stable-diffusion-webui-forge repository as a fork/platform on top of AUTOMATIC1111's Stable Diffusion WebUI, establishing Forge as a from-day-one derivative of the existing WebUI ecosystem rather than an independent project. - 2February 5, 2024 β First packaged release
Why it matters: The GitHub Releases page's "latest" tag publishes a downloadable, pre-packaged Windows build, giving users an installer-style option alongside the manual git-clone workflow. - 3July 22, 2024 β "previous_versions" release published
Why it matters: Preserves older installer builds as Forge's packaged download continued to change, so users on older CUDA/PyTorch combinations still had a working build to fall back on. - 4September 7, 2024 β Flux Realistic sampler added
Why it matters: Per the project's own NEWS.md, a new sampler tuned for Flux models shipped alongside Forge's built-in Flux support via NF4/GGUF quantization. - 5October 28, 2024 β "sd35" branch work begins (PR #2183)
Why it matters: Per NEWS.md, work started on a branch evaluating Stable Diffusion 3.5 support, quantization methods, and sampling techniques ahead of a possible merge into the main branch. - 6July 31, 2025 β Most recent code push (per GitHub API)
Why it matters: The newest commit activity this review could confirm on the main repository as of its publish date β useful context for judging current development pace before relying on Forge for a workflow that needs frequent updates.
What Can You Do With Forge?
Forge's feature set centers on changing how the WebUI manages GPU memory and executes models, while adding several capabilities as built-in features that require separate extensions in the original AUTOMATIC1111 WebUI. Here is what each part actually does, per Forge's own GitHub README.
- Rebuilt UNet-patching backend β Forge's README describes a backend rebuild for "unet patching" targeting lower GPU memory use and faster inference speed than the original WebUI, on the same checkpoint and hardware
- Built-in ControlNet and IP-Adapter support β full support for ControlNets and IP-Adapters ships as part of the app itself, rather than needing the separate ControlNet extension the original WebUI requires
- LoRA support β compatible with LoRA models, including low-bit/quantized variants per the README
- Flux model support β supports Flux checkpoints using BitsandBytes NF4 and GGUF quantization formats, letting large Flux models run on more modest GPUs than an unquantized version would need
- Full preprocessor suite β the preprocessing tools ControlNet relies on (edge detection, depth maps, pose estimation, and similar) are included rather than requiring separate downloads
- Gradio 4 interface with a canvas tool β a modernized UI compared to the original WebUI's older Gradio version, including built-in canvas functionality for inpainting-style workflows
- Extension compatibility β Forge is designed to work with much of the original AUTOMATIC1111 WebUI's extension ecosystem (the README names extensions such as Adetailer as working examples), though not every extension is guaranteed compatible
Usage Examples: Three Ways to Use Forge
These are concrete workflows built from Forge's documented features above, not hypothetical use cases.
Install Stable Diffusion WebUI Forge
Stable Diffusion WebUI Forge is available on Windows, macOS, and Linux, as a one-click packaged build on Windows or a manual git-clone setup on any platform. Links below are from the official GitHub repository β always verify against that page directly, since installer build URLs can change between releases.
Windows (packaged build)
- Install Method:
- Get the CUDA+PyTorch package from GitHub Releases, then run
update.batandrun.bat.
macOS / Linux (clone + script)
- Install Method:
- Clone the repository, then run
webui.shto set up a Python venv and launch Forge.
Windows (advanced / manual)
- Install Method:
- Clone the repository and run
webui-user.batfor a manual Python setup.
There is no single official .exe or .dmg installer β Windows users typically use the packaged build above, while macOS and Linux users set up Forge via git clone plus a Python virtual environment, the same pattern used by the original AUTOMATIC1111 WebUI. A GPU is strongly recommended for practical generation speed; Forge itself documents no single fixed VRAM minimum, since actual requirements depend on the checkpoint (standard SD 1.5/SDXL vs. the larger Flux family) and the output resolution you choose.
Forge Pricing: Is It Really Free?
Yes β Forge has no paid tier. There is no pricing page, storefront, or account system tied to the GitHub repository, and the AGPL-3.0 license applies to the entire application.
- No subscription, no paid tier, no usage limits imposed by Forge itself
- No account or sign-up required to install or run the app
- The application is licensed AGPL-3.0, a copyleft license β unlike a permissive MIT/Apache license, distributing a modified version (including running it as a network service) generally requires releasing your modified source code under the same license; read the full license text yourself if this matters for your use case
- The only real costs are indirect: the GPU hardware needed to run it well, and electricity β Forge does not charge for the software itself or for models you download separately from HuggingFace or CivitAI
Forge vs. AUTOMATIC1111 WebUI
Forge and AUTOMATIC1111's Stable Diffusion WebUI share the same lineage β Forge is a fork built directly on top of it β so this is less "different app" and more "same app, rebuilt backend." The clearest differences are in memory management, built-in features, and how actively each project currently ships updates.
Relationship
- Forge:
- Fork built on top of AUTOMATIC1111 WebUI
- AUTOMATIC1111 WebUI:
- The original, more widely extended project
Backend / memory management
- Forge:
- Rebuilt UNet-patching backend for lower VRAM use, faster inference
- AUTOMATIC1111 WebUI:
- Original backend; VRAM tricks via flags/extensions
ControlNet / IP-Adapter
- Forge:
- Built in, no extension needed
- AUTOMATIC1111 WebUI:
- Requires the separate ControlNet extension
Flux model support
- Forge:
- Built in, with NF4/GGUF quantization
- AUTOMATIC1111 WebUI:
- Depends on community extensions, not native
Extension ecosystem
- Forge:
- Compatible with much of it, but narrower
- AUTOMATIC1111 WebUI:
- The largest, most mature SD extension ecosystem
Recent commit activity
- Forge:
- Most recent push found: July 31, 2025
- AUTOMATIC1111 WebUI:
- Check the project's own commit history directly
If your priority is faster inference and lower VRAM use with ControlNet and Flux support out of the box, Forge's rebuilt backend is the reason to choose it. If your priority is the widest possible extension ecosystem, or you rely on an extension Forge does not yet support, the original AUTOMATIC1111 WebUI is worth evaluating directly at its GitHub repository. Both are free, open-source, and installed the same way (packaged Windows build or git clone); verify current compatibility for any specific extension you depend on before switching.
Who Should Use Forge?
Whether Forge fits depends on whether faster inference and lower VRAM use matter more to you than the widest possible extension ecosystem or guaranteed frequent updates.
Forge vs. Other Local Image Generation Tools
Forge is one of several local, open-source interfaces for Stable Diffusion and related image models. Here is how it sits next to other options in that space β see the Local LLM Software Directory for the full catalog, and the dedicated Forge vs. AUTOMATIC1111 WebUI comparison above for the closest head-to-head.
- AUTOMATIC1111 Stable Diffusion WebUI β the original project Forge is built on top of, with the largest extension ecosystem; see the dedicated comparison section above and the AUTOMATIC1111 review for full details.
- ComfyUI β a node-based interface for the same underlying Stable Diffusion/Flux models, trading Forge's more familiar tab-based UI for a visual workflow graph with finer low-level control; see the ComfyUI review.
- Fooocus β also created by lllyasviel, but built for the opposite goal: a simplified, opinionated interface aimed at beginners who want good results with minimal parameter tuning, rather than Forge's full WebUI parameter set; see the Fooocus review.
- InvokeAI β another polished, actively developed local Stable Diffusion UI with a strong focus on a unified canvas workflow and a more curated feature set than the WebUI-derived tools; see the InvokeAI review.
Common Mistakes When Evaluating Forge
Most confusion about Forge comes from its relationship to AUTOMATIC1111's WebUI, its extension compatibility, or assuming it has a fixed hardware requirement.
Frequently Asked Questions
What is Stable Diffusion WebUI Forge?
Stable Diffusion WebUI Forge (github.com/lllyasviel/stable-diffusion-webui-forge) is a free, open-source local image-generation web app, built as a fork on top of AUTOMATIC1111's Stable Diffusion WebUI by lllyasviel, the creator of ControlNet, with a rebuilt backend aimed at faster inference and lower VRAM use.
Is Stable Diffusion WebUI Forge free?
Yes. There is no pricing page or paid tier, and the AGPL-3.0 license applies to the entire application. If you use paid HuggingFace or CivitAI model hosting, or need a more powerful GPU, those are separate costs unrelated to Forge itself.
Is Forge open source? What license does it use?
Yes, Forge is open source under the AGPL-3.0 license, per the repository's LICENSE.txt file. AGPL-3.0 is a copyleft license: distributing a modified version, including as a network service, generally requires releasing the modified source under the same license β read the license text yourself for any legal decision.
Is Forge the same as AUTOMATIC1111's Stable Diffusion WebUI?
No, but it is closely related. Forge is a fork built directly on top of AUTOMATIC1111's Stable Diffusion WebUI, reusing much of its checkpoint format and extension API while rebuilding the backend that manages GPU memory and model execution for lower VRAM use and faster inference.
Does Forge require a GPU?
A GPU is strongly recommended for practical use. Forge documents no single fixed VRAM minimum β actual requirements depend on the checkpoint you load (standard SD 1.5/SDXL vs. the larger Flux family) and your chosen output resolution.
Does Forge support Flux models?
Yes. Forge supports Flux checkpoints using BitsandBytes NF4 and GGUF quantization formats, which lets larger Flux models run on GPUs that could not hold an unquantized version.
Does Forge include ControlNet?
Yes, built in. Forge ships full ControlNet and IP-Adapter support along with the preprocessing tools they rely on, so you do not need to install the separate ControlNet extension the original AUTOMATIC1111 WebUI requires.
What platforms does Forge support?
Windows, macOS, and Linux. Windows has a packaged, one-click build available from GitHub Releases; macOS and Linux are set up via a manual git-clone-and-script install, the same pattern the original AUTOMATIC1111 WebUI uses.
Who created Stable Diffusion WebUI Forge?
Forge was created by lllyasviel (Lyumin Zhang), who is also the creator of ControlNet and of Fooocus, another local Stable Diffusion interface.
Is Forge still actively maintained?
Development has slowed compared to Forge's first year. GitHub's API shows the repository's most recent code push was July 31, 2025, as of this review. The repository was not archived and remained publicly listed β check its commit history directly for the current status before relying on frequent updates.