🎨 FLUX.2 [klein] 4B AIO Workflow v1

Official workflow for FLUX.2 [klein] 4B AIO models

✨ What's Included

• ⚡ Optimized Text-to-Image generation

• 🔼 Optional upscaling pipeline

• 💾 Auto-save with full metadata

• 🔍 Before/After image comparison

• 📖 Built-in LoRA support (toggle on/off)

• 📐 Pre-configured for 1024×1024 native resolution

🔄 Compatible Models

🟡 FP8-AIO (~7.7 GB) – Recommended

🔵 FP16-AIO (~15 GB) – Older GPUs

🟢 BF16-AIO (~15 GB) – Best Quality

All versions work with this workflow!

⚙️ Pre-Configured Settings

• Steps: 4 (step-distilled)

• CFG: 1.0 ⚠️

• Sampler: euler

• Scheduler: simple

• Resolution: 1024×1024

⚠️ CRITICAL: Keep CFG at 1.0!

This is a distilled model. Higher CFG = worse results!

📦 Required Custom Nodes

Install via ComfyUI Manager:

• rgthree-comfy – Image comparison, LoRA toggle

• comfyui_image_metadata_extension – Auto-save metadata

🚀 Quick Start

1. Download workflow (.json)

2. Import into ComfyUI

3. Select your FLUX.2 [klein] 4B AIO model

4. Write your prompt

5. Generate!

📂 Model Installation

ComfyUI/models/checkpoints/
└── flux-2-klein-4b-fp8-aio.safetensors

💡 Workflow Features

🎨 Base Generation

• Natural language prompts

• 4-6 steps for best results

• No negative prompts needed

🔼 Upscaling (Optional)

• 1.5x scale by default

• Denoise 0.3 preserves details

• Adjustable to your needs

💾 Auto-Save

• PNG with full metadata

• Easy CivitAI uploads

• All settings preserved

🔍 Image Comparison

• Side-by-side view

• Base vs Upscaled

• Slide comparison mode

🎯 Tips

• Portrait: Set resolution to 768×1024 or 832×1216

• Landscape: Set resolution to 1024×768 or 1216×832

• Square: Keep 1024×1024 (native, best quality)

• LoRAs: Toggle via "Fast Groups Bypasser" node

🔗 Links

• Model Download: CivitAI Model Page

• Original Model: Black Forest Labs

• rgthree-comfy: GitHub

• Metadata Extension: GitHub

Version: 1.0
Updated: January 2026
License: Apache 2.0

Sub-second image generation made simple! ⚡