Image-to-Image Dataset Creation Workflow EP -2
Create high-quality Image-to-Image datasets using ComfyUI with prompt automation, latent conditioning, LoRA control, and multi-stage refinement. Includes reference-based generation + upscale pipeline for consistent dataset outputs. Beginner-friendly Hindi workflow.
Beginner Friendly
ComfyUI Hindi Series
Consistent Images
Dataset Creation
image to image
Image to Image Dataset
one image to more image
Upscaling
2
214
Nodes & Models
UNETLoader
flux-2-klein-9b-fp8.safetensors
CLIPLoader
qwen_3_8b_fp8mixed.safetensors
VAELoader
flux2-vae.safetensors
UpscaleModelLoader
EmptyFlux2LatentImage
LoadImage
Note
LoraLoaderModelOnly
klein_slider_anatomy-image-to-image-dataset-c-oyY4tPve.safetensors
Realism_Engine_Klein_V2-image-to-image-dataset-c-oyY4tPve.safetensors
CLIPTextEncode
Reroute
ImageScaleToTotalPixels
VAEEncode
ReferenceLatent
PreviewImage
SaveImage
ImageUpscaleWithModel
ImageScaleBy
KSampler
VAEDecode
Image Comparer (rgthree)
CR Prompt List
KSampler (Efficient)
KSampler (Efficient)
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
⚙️ 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
Read more
