Stable Diffusion WebUI Forge is a fork of AUTOMATIC1111 by lllyasviel, the author of ControlNet and Fooocus. It keeps the familiar interface, prompts and most extensions, but replaces the engine underneath with a memory manager that is far more efficient. Forge added Flux support before most other classic UIs and can run Flux on 6 to 8 GB cards with the right quantized files.

Pick Forge if you know and like the A1111 layout but want Flux, SDXL on a small graphics card, or simply faster generation. It is Windows and Linux only in practice. If you want video models or exotic workflows, use ComfyUI.

Sem instalação: rode modelos de IA na nuvem

BitVector Prism é a forma mais fácil de começar: escolha um modelo, digite um prompt e gere em um app web limpo, sem configurar nada. O BitVector também está disponível no Discord e na web (SpyGlass).

Experimentar o BitVector

O que você precisa

Windows 10 or 11 or Linux with an NVIDIA graphics card; 6 GB VRAM is enough for SDXL and for Flux in NF4 or GGUF form, 12 GB or more for full Flux. 16 GB of system RAM is recommended because Forge swaps model parts to RAM. Plan 20 GB of disk space plus your models.

Como obter e instalar o Stable Diffusion WebUI Forge

Download the one-click package (webui_forge_cuXXX_torchXXX.7z) from the GitHub releases, extract it with 7-Zip to a folder without spaces, and run update.bat once.
Start run.bat. The first run installs the remaining pieces and opens the interface in your browser at http://127.0.0.1:7860.
Manual install: clone the repository and run webui-user.bat (Windows) or webui.sh (Linux), as with A1111.
Place checkpoints in models/Stable-diffusion, Flux unet files in models/Stable-diffusion as well (or models/unet), the Flux text encoders and VAE in models/text_encoder and models/VAE, and LoRAs in models/Lora.
If you already have A1111, do not copy its folder over Forge; instead point Forge at your existing models through the webui-user.bat argument --ckpt-dir or by editing the model paths in the settings.

Baixe apenas pelos links oficiais desta página. Instaladores de outros lugares podem estar desatualizados ou adulterados. O ModelTrends.ai não hospeda arquivos.

Sem instalação: rode modelos de IA na nuvem

O PirateDiffusion é exclusivo do Telegram e feito para profissionais: milhares de modelos, LoRAs e workflows por comandos de chat, com geração ilimitada a preço fixo.

Experimentar o PirateDiffusion

Como usar o Stable Diffusion WebUI Forge

At the top, choose the UI preset: sd, xl or flux. It shows the right options for that model family.
For Flux, select the unet file as the checkpoint and tick the matching VAE / text encoder files in the dropdown next to it (ae, clip_l and t5xxl). Set Distilled CFG to about 3.5 and CFG to 1.
Write your prompt, set the size (1024x1024 is a good start for SDXL and Flux) and the steps (20 for Flux dev, 4 for schnell, 25 for SDXL), then press Generate.
LoRAs work as in A1111: click one in the Lora tab to add <lora:name:1> to the prompt and include its trigger words from the model page.
Results are saved in the outputs folder. Use the GPU Weights slider at the top if you run out of memory: a lower value keeps more of the model in system RAM.

Dicas e problemas comuns

Flux NF4 and GGUF Q4/Q5 files are the best choice for 8 GB cards; they look almost identical to the full model and load in a fraction of the memory.
Many A1111 extensions work in Forge, but not all; check the Forge discussions page before installing an extension that changes the sampling code.
Keep Forge updated with update.bat; model support arrives in updates, not in extensions.
If generation is slow, make sure the GPU Weights value is not set too low; too much swapping to RAM costs speed.
Settings and prompt styles can be copied from A1111: the ui-config.json and styles.csv files are compatible.

Modelos que você pode rodar no Stable Diffusion WebUI Forge

Forge runs Stable Diffusion 1.5, SDXL (including Pony and Illustrious variants) and Flux checkpoints and LoRAs from this catalog. For Wan, Hunyuan and other video models, use ComfyUI or Wan2GP instead.

Modelos Flux

Modelos Stable Diffusion XL 1.0

Modelos Stable Diffusion 1.5

Sem instalação: rode modelos de IA na nuvem

Sem GPU, sem downloads, sem Python. Esses serviços rodam os mesmos modelos de código aberto em seus próprios servidores, então você começa a gerar pelo navegador, Discord ou Telegram em um minuto.

Experimentar o BitVector

Experimentar o PirateDiffusion

Links

Estes links levam aos sites dos próprios criadores. O ModelTrends.ai não é afiliado a eles e não hospeda software.

Outros apps locais

Como instalar o ComfyUI

The node-based powerhouse for Stable Diffusion, Flux, Wan video and almost every new open source model. Free, open source, for Windows, Mac and Linux.