LlamaIndex
LlamaIndex builds document parsing and extraction software plus open-source tools for AI applications, helping teams turn enterprise documents into usable data for AI agents and workflows.

Stable Diffusion WebUI Forge is a platform built on top of Stable Diffusion WebUI that optimizes resource management, speeds up inference, and makes experimental features easier to study. It targets users who run Stable Diffusion locally and want better performance and memory handling without abando
Stable Diffusion WebUI Forge is a platform built on top of Stable Diffusion WebUI, itself based on Gradio, that aims to make development easier, optimize resource management, speed up inference, and provide a place to study experimental features. It is aimed at people who already use the AUTOMATIC1111 WebUI and want more performance and flexibility from the same interface.
The name "Forge" is borrowed from Minecraft Forge, and the project's stated goal is to become SD WebUI's Forge. Rather than replacing the original interface, Forge layers its own optimizations and experiments on top of it. That approach matters for users with existing workflows, extensions, and habits: the familiar tabs and controls remain, while the underlying resource handling and inference paths change.
Forge is currently based on SD-WebUI 1.10.1 at a specific commit. Because the original SD-WebUI is described as almost static now, Forge syncs with it roughly every 90 days, or sooner when important fixes arrive. That cadence tells you what to expect: Forge is not chasing every upstream change, but it does not abandon upstream maintenance either. Users who depend on a specific WebUI behavior should note the base version when troubleshooting.
Forge natively supports Flux BitsandBytes NF4 and GGUF Q8_0, Q5_0, Q5_1, Q4_0, and Q4_1 models. These are exposed through a GPU weight slider, a Queue/Async Swap toggle, and a swap location toggle. LoRA support is available for Flux BNB NF4 and GGUF Q8_0, Q5_0, and Q4_0 models. The project's own guidance is blunt about tuning: do not set the GPU weight too high, because lowering it solves the vast majority of reported Flux performance problems.
The recommended route is the one-click installation package, which bundles git and Python. The recommended build pairs CUDA 12.1 with Pytorch 2.3.1. A CUDA 12.4 + Pytorch 2.4 build is described as the fastest, but MSVC may be broken and xformers may not work. A CUDA 12.1 + Pytorch 2.1 build preserves the previously used older environments. After downloading and uncompressing, you run update.bat to update and run.bat to launch. The project stresses that running update.bat matters, since skipping it can leave you on a previous version with unfixed bugs.
Forge integrates ControlNet and FreeU, and includes LayerDiffuse for transparent image editing. FreeU is exposed as a script with sliders for B1, B2, S1, and S2, and it patches the UNet before sampling. The project also maintains a Forge extension list and an extension replacement list, which helps users map existing WebUI extensions onto Forge equivalents. Community discussions cover LoRA precision on low-bit models, skipping LoRA patching, loading a LoRA once instead of every generation, and reporting LoRAs that do not work.
The Gradio 4 UI has one notable quirk: you need to use the right mouse button to move the canvas. The project maintains discussion threads for common failures, including "Connection errored out" and "Press anykey to continue" messages, as well as saving Flux BitsandBytes UNet or checkpoint files. A soft advertisement removal policy is also documented.
WebUI Forge is under construction, and the project warns that docs, UI, and functionality may change with updates. That is worth weighing if you need a stable, frozen environment. For users who want to follow Flux and other experimental features as they develop, the tradeoff is a moving target in exchange for earlier access to optimizations and integrations.
Forge keeps the A1111 interface almost exactly as you know it — the difference is underneath. Where A1111 loads a model and hopes your GPU can hold it, Forge manages GPU memory dynamically: its rewritten backend moves model weights between VRAM and system RAM as needed, controlled by a "GPU Weight" slider, instead of forcing the whole model to stay resident on the card. That single change is why the two programs behave so differently on the same hardware.
In practice, independent benchmarks consistently put Forge roughly 30% or more ahead of A1111 on SDXL generation, with the gap widening as more LoRAs are added to a prompt. The VRAM story matters even more. A 1024×1024 SDXL render that pushes an 8 GB card into out-of-memory errors in A1111 typically runs in Forge with room to spare, and SD 1.5 models are comfortable on 4–6 GB cards. Forge also runs models A1111 cannot handle natively at all: Flux.1 is supported out of the box in quantized formats — BitsandBytes NF4 and GGUF builds — which bring a 12-billion-parameter model down to a size that fits on 6–8 GB of VRAM, with LoRA support included.
So which should you install? If you already use A1111, switching is low-risk: Forge can be installed alongside it and can reuse your existing checkpoints and most extensions, and the one-click package (CUDA 12.1 with PyTorch 2.3.1 is the recommended build) needs no Python setup. Stay with A1111 only if your workflow depends on a niche extension that has not been ported yet. For everyone else — and especially for anyone on an 8 GB card who wants SDXL or Flux — Forge is the same familiar tool, simply faster and far less likely to run out of memory.
What is SD WebUI Forge? Stable Diffusion WebUI Forge is a platform built on top of the original Stable Diffusion WebUI (A1111) by developer lllyasviel, the creator of ControlNet. It keeps the familiar A1111 interface but rewrites resource management underneath, so inference runs faster and uses GPU memory more efficiently.
Is Forge faster than A1111? Yes. Independent benchmarks typically show Forge generating SDXL images around 30% faster than A1111 on the same hardware, and it avoids many of the out-of-memory errors A1111 hits on 8 GB graphics cards, because model weights can be offloaded dynamically instead of living entirely in VRAM.
Can SD WebUI Forge run Flux models? Yes — this is one of its biggest advantages over A1111. Forge natively supports Flux.1 in BitsandBytes NF4 and GGUF quantized formats, which reduce the model enough to run on roughly 6–8 GB of VRAM, and those quantized builds still support LoRAs.
Is Stable Diffusion WebUI Forge free? Yes. Forge is open source under the AGPL-3.0 license and free to download from GitHub, either as a one-click package with Git and Python included or as a standard git clone install.
Subscription pricing is not listed.
Features, pricing and availability may change. Always verify details on the official website. Maintained by steven · Last updated Nov 10, 2026 →
LlamaIndex builds document parsing and extraction software plus open-source tools for AI applications, helping teams turn enterprise documents into usable data for AI agents and workflows.
Flowith is an agentic AI workspace that combines a whiteboard-style canvas, access to 40+ image and video models, and mini apps for design, e-commerce, and social content.
Fine-tune open models faster with far less memory
Groq is an AI inference cloud for developers and enterprises that need low-latency model responses at scale, delivered through custom LPU hardware and a tiered platform spanning bare metal, inference, and enterprise controls.