Install FLUX in Stable Diffusion Forge: Setup Guide

This guide shows you how to install and run FLUX models in Stable Diffusion Forge. It’s based on a working setup guide from Reddit and verified installation steps I’ve tested myself. The instructions are simple and written for beginners — no technical jargon, just clear steps.

Source reference: Reddit: Getting started with FLUX in Forge

Requirements

Before you start, make sure you have:

FLUX works best on NVIDIA GPUs — that’s the honest truth. AMD support is experimental, so your mileage may vary.

Step 1: Install Stable Diffusion Forge

1. Download Forge from the official GitHub: lllyasviel/stable-diffusion-webui-forge
2. Click Code → Download ZIP or clone the repo.
3. Extract it to a folder like:

   1
   C:\AI\Forge

4. Run the file run.bat (Windows) or run.sh (Linux).
5. Wait until Forge finishes installing dependencies — this might take a few minutes, so grab some coffee.
6. Open the interface at:

   1
   http://127.0.0.1:7860

Step 2: Download FLUX Model

Download FLUX from an official source:

• FLUX.1-dev (high quality): FLUX.1-dev
• FLUX.1-schnell (fast): FLUX.1-schnell

Place the .safetensors model file here:

1
Forge/models/Stable-diffusion/

Step 3: Required Text Encoders (Correct Files & Paths)

FLUX in Forge requires two text encoders:

1. CLIP-L (ViT-L/14) – clip_l.safetensors
2. T5-XXL (v1.1) – choose one of:
   • t5xxl_fp16.safetensors (~9.8 GB, best quality; needs ~32 GB system RAM)
   • t5xxl_fp8_e4m3fn.safetensors (~4.9 GB, lighter; recommended for most PCs)

Download (trusted source):

• Hugging Face: comfyanonymous/flux_text_encoders

Put both files here (create folder if missing):

1
Forge/models/text_encoder/

Depending on your installation, the path may be webui/models/text_encoder/ inside the Forge directory.

You do not need to download OpenAI’s CLIP repository or Google’s full T5 model separately — the .safetensors files above are all you need. Don’t overthink this part.

Step 4: Enable FLUX in Forge

1. Start Forge.
2. Go to Settings → Stable Diffusion.
3. Enable FLUX compatibility mode (if available).
4. Select your model:
   • flux1-dev-fp16.safetensors or flux1-schnell-fp16.safetensors
5. Click Apply settings.
6. Reload UI.

Step 5: Recommended Settings

For low VRAM GPUs, enable:

• Settings → Optimization → Low VRAM Mode
• Use FLUX.1-schnell for faster generation

Troubleshooting

🔧 Optional: Enable LoRA Support for FLUX in Forge

Forge supports LoRA for FLUX, but you must enable it properly.

Step 1: Download FLUX LoRA

You can find FLUX-ready LoRAs on:

• Civitai
• Hugging Face

Place LoRA files in:

1
Forge/models/Lora/

Step 2: Enable LoRA Loader

1. Go to Forge → Extensions
2. Install Lora Block Weight (optional, improves LoRA control)
3. Restart Forge

Step 3: Use LoRA in Prompt

In the prompt box, call LoRA like this:

1
<lora:your_flux_lora:0.8>

Adjust strength between 0.6 – 1.0.

🧩 Add ControlNet for FLUX in Forge

Forge supports ControlNet for pose, depth, and edge guidance.

Install ControlNet Extension

1. Go to Extensions → Available
2. Search ControlNet
3. Install and restart UI

Download ControlNet Models

Place models in:

1
Forge/models/ControlNet/

Recommended:

Enable ControlNet

1. Check Enable ControlNet
2. Upload reference image
3. Generate with guidance

⚙️ VRAM Optimization for FLUX in Forge

If you have a low VRAM GPU (8–12GB), apply these settings:

• Settings → Optimization → Enable low VRAM mode
• Use batch size = 1
• Use FLUX.1-schnell instead of FLUX-dev
• Reduce resolution: 832×832 or 768×1024
• Disable VAE tiling unless needed

🔼 Install High-Quality Upscalers (Optional)

To improve final image quality, install a 4x AI upscaler.

Recommended Upscalers

Place upscalers here:

1
Forge/models/ESRGAN/

Enable in Forge under Extras → Upscaling.

✅ Recommended Generation Settings for FLUX in Forge

These settings provide a good starting point for most users:

💡 Prompt Examples for FLUX

General Quality Prompt

1
hyperrealistic cinematic portrait, dramatic light, depth of field, masterpiece quality, 85mm lens

Landscape Prompt

1
epic mountain valley, soft fog, sunrise glow, ultra detailed, atmospheric depth, realistic lighting

Cyberpunk Prompt

1
futuristic neon city street at night, rain reflections, cinematic film still, volumetric lighting

Use a negative prompt for best results:

1
blurry, lowres, watermark, text, distorted, bad hands, noisy

🚀 Performance Tips for Forge + FLUX

• Disable VAE Precision High to save VRAM
• Use xformers if available
• Close browser tabs while generating
• Avoid CFG > 6 (FLUX prefers lower values)
• Use schnell model for faster image previews

🛠️ Troubleshooting

Related Guides

• FLUX in ComfyUI: /blog/flux-comfyui-guide
• SDXL Best Practices: /blog/sdxl-best-practices-guide
• Stable Diffusion Prompting: /blog/stable-diffusion-prompting-guide
• Ostris AI Toolkit (LoRA training): /blog/ai-toolkit-guide