🔮 Transform Your Images with Natural Language Instructions

The Perfect Companion for Seamless Editing

This guide outlines a complete workflow using FluxFill-Dev with ICEdit-MoE-LoRA for powerful natural language-based image editing.

📝 Overview: The Future of Image Editing is Here

ICEdit-MoE-LoRA represents a breakthrough in AI image editing - a lightweight Low-Rank Adaptation (LoRA) module that brings professional-grade, instruction-based editing capabilities to large diffusion transformers. By intelligently fine-tuning only 1% of parameters using just 0.5% of the original training data, it achieves extraordinary editing capabilities that match or even surpass premium commercial solutions.

📌 Complete Workflow

This workflow combines FluxFill + ICEdit-MoE-LoRA for editing images with natural language instructions. For enhanced results, diffusion models like LEOPARD , Bernoulli or SPEED_Q8 with the Gemini auto prompt node in 4-8 steps.(i used very light flux model which need less steps)

A diptych with two side-by-side images of the same scene. On the right, the scene is the same as on the left but "your edit"(prostyle included just put ur edit)

🛠️ Complete Installation Guide

📁 Setup Structure

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   ├── (basic)📄 fluxfill-dev.safetensors # https://civitai.com/models/1108146/fluxfill-devgguf
│   │   ├──(option1)📄 LEOPARD.gguf     # https://civitai.com/models/1534861
│   │   ├──(option2)📄 Bernoulli.gguf  # https://civitai.com/models/682369
│   │   └──(option3)📄 SPEED_Q8.gguf   # https://civitai.com/models/661102/speedq8
│   ├── 📂 text_encoders/
│   │   ├── (basic)📄 clip_l.safetensors
│   │   ├── (option1)📄 t5xxl_fp16.safetensors
│   │   ├── (option2)📄 t5xxl_fp8_e4m3fn.safetensors
│   │   └── (option3)📄 t5xxl_fp8_e4m3fn_scaled.safetensors
│   ├── 📂 vae/
│   │   └── 📄 ae.safetensors
│   └── 📂 lora/
│       └── 📄 ICEdit-MoE.safetensors

💎 Essential Components

✨ LoRA Models - The Magic Touch

• ICEdit-MoE.safetensors - The star of the show!

🔤 Text Encoders - The Brain Behind Natural Language Understanding

Note: You only need to choose ONE of the T5XXL options below based on your hardware capabilities

• clip_l.safetensors - Core understanding model

• T5XXL Options (choose only one):

  • t5xxl_fp16.safetensors - Highest quality, more demanding

  • t5xxl_fp8_e4m3fn.safetensors - Balanced performance

  • t5xxl_fp8_e4m3fn_scaled.safetensors - Optimized for speed

🎭 VAE - The Visual Artist

• ae.safetensors - Included in the FluxFill package

🖼️ Diffusion Models - The Imaging Engine

• fluxfill-dev.safetensors - Core diffusion model for editing

##(You only need to choose ONE of those or use flux-dev itself its up to u)

• LEOPARD.safetensors - Advanced diffusion model for refined outputs

• Bernoulli.safetensors - Alternative diffusion model for enhanced generation

• SPEED_Q8.gguf - Enhanced flux-dev model for improved results

Note: Any of these diffusion models can be used for efficient editing with fewer steps (4-8 steps recommended). Use flux-dev-fill for editing and flux-dev for enhancing results.

🎬 Example Instructions You Can Try

• "Add a majestic castle on the mountain in the background"

• "Change the color of the car from red to electric blue"

• "Remove the person from the image and replace with a flowering tree"

• "Transform the style to watercolor painting while maintaining all details"

• "Add dramatic sunset lighting to the scene"

👨‍💻 Developer Information

This workflow guide was created by Abdallah Al-Swaiti:

• LinkedIn Profile

• GitHub

• Hugging Face

For additional tools and updates, check out the OllamaGemini Node: GitHub Repository

🙏 Community Recognition

A special thank you to sanaka87 for revolutionizing open-source image editing by releasing the ICEdit-MoE-LoRA checkpoint and demos. Your contribution empowers creators everywhere and accelerates innovation across the AI art community!