Image-to-Image Dataset Creation Workflow EP -2

Image-to-Image Dataset Creation Workflow EP -2

This workflow is designed to create consistent and diverse Image-to-Image datasets using ComfyUI. It combines prompt generation + reference latent conditioning + LoRA + multi-stage refinement + upscaling to produce dataset-ready outputs.

🎥 YouTube Tutorial

(Add your video link here)

⚙️ How this workflow works:

1️⃣ Prompt Automation

→ Uses CR Prompt List to generate multiple variations
→ Ensures dataset diversity with consistent structure

2️⃣ Image-to-Image Conditioning

→ Input images are encoded into latent space
→ ReferenceLatent keeps composition consistent

3️⃣ Controlled Generation

→ CLIP prompts guide transformation
→ LoRA adds specific style/anatomy control

4️⃣ Latent Refinement

→ KSampler improves details and realism
→ Better lighting, texture, and consistency

5️⃣ Upscaling Pipeline

→ Images are enhanced using upscale models
→ Final outputs are sharper and dataset-ready

✨ Key Features

• Consistent Image-to-Image dataset generation
• Prompt-based automation for bulk creation
• Reference-based composition control
• High-quality outputs with refinement
• Built-in upscale pipeline
• Clean and scalable workflow

🎯 Best Use Cases

• Dataset creation for training models
• Image-to-Image fine-tuning datasets
• AI influencer dataset generation
• Style transfer datasets
• Research & experimentation

🛠️ Requirements

• ComfyUI installed
• Required models (UNET, VAE, LoRA)
• GPU recommended

📌 Note

This workflow is part of my ComfyUI Hindi Series