Have a ComfyUI workflow? Get a readable Workflow Doc in your browser.Make a workflow readable.Try it now ↗
ComfyUI LoRA Training Prep Workflow (with Flux1.d) This ComfyUI workflow is built to simplify dataset preparation for LoRA training by automating batch image upscaling and annotation with the power of Flux1.d . It...
Runtime profile
Source description
ComfyUI LoRA Training Prep Workflow (with Flux1.d) This ComfyUI workflow is built to simplify dataset preparation for LoRA training by automating batch image upscaling and annotation with the power of Flux1.d . It processes entire folders of images, leveraging Flux1.d’s advanced image generation and enhancement capabilities to upscale images with exceptional clarity and detail. After upscaling, it runs automated tagging/annotation to produce descriptive captions, saving them alongside each image for seamless LoRA training integration.
Key features:
Flux1.d Image Enhancement: Utilizes Flux1.d to upscale and refine images, ensuring high-resolution, high-fidelity training data.
Batch Processing: Handles large image sets with ease, making it ideal for dataset creation at scale.
AI-generated commentary
AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.
Batch Upscale and Annotator Workflow is for preparing LoRA training datasets from folders of images: in ComfyUI, it batch-upscales images with Flux1.d, generates captions or tags through Florence, and saves paired image and text files.
It processes image folders, upscales the images, then runs automated tagging or annotation and saves captions alongside the images.
Input: a folder of images for batch processing.
Outputs: upscaled images with descriptive captions or tags, saved as paired image and text files for LoRA training.
Use it in ComfyUI with drag-and-drop and modular nodes; the visual layout can be customized.
The listed model files and model are: flux1-dev-fp8.safetensors, t5xxl_fp16.safetensors, ae.safetensors, clip_l.safetensors, 4x_NMKD-Superscale-SP_178000_G.pth, and microsoft/Florence-2-base.
The listed ComfyUI node packs are: comfyui_ultimatesdupscale, comfyui-custom-scripts, comfyui-florence2, comfyui-kjnodes, rgthree-comfy, was-node-suite-comfyui, and was-ns.
The base model is Flux.1 D.
The annotation stage uses Florence, and the modular node layout can be customized.
Before use, verify that ae.safetensors and was-ns are available under the exact names shown.
Suggestion · not verified
Before use, verify which ComfyUI node packs to use for comfyui-florence2 and was-node-suite-comfyui.
Suggestion · not verified
Does this need to be edited?
Sign in to send an edit request.
Sources
1 sourceSource excerpts
1 excerptSource context: 410 downloads · Type Workflows · Base model Flux.1 D
Automated Annotation: Generates captions or tags for each image (via Florence), drastically reducing manual labeling effort.
LoRA-Ready Output: Outputs neatly paired image and text files in a structure optimized for LoRA training workflows.
Fully ComfyUI-Integrated: Drag-and-drop functionality, modular node setup, and a clear visual workflow make it easy to customize.
This workflow is perfect for AI creators and LoRA trainers who want high-quality, consistent datasets while minimizing prep time.
Estimated VRAM requirement
Estimate unavailable
20.9 GB across 4 of 7 model files. Model file total + 25% loading overhead + 2 GB execution buffer, rounded up.
Requirements
13 requirementsUpscaler · 63.9 MB · PT · Unknown
VAE · Unknown
Text encoder · 235 MB · SAFETENSORS · Unknown
Upscaler · Registry · ComfyUI_UltimateSDUpscale
Checkpoint · 11.5 GB · SAFETENSORS · Unknown
Vision model · Hugging Face · microsoft/Florence-2-base
Text encoder · 9.12 GB · SAFETENSORS · Unknown
Node pack · Registry
comfyui-florence2
PossibleNode pack · Registry
Node pack · Registry
Node pack · Registry
was-node-suite-comfyui
PossibleNode pack · Registry
was-ns
Not resolvedNode pack · Registry
