Train a custom LoRA on Civitai

Matt Hallett Visual·Jan 2026

Demo summary

A step-by-step walkthrough of uploading a dataset of streetcar images to Civitai, using auto-labeling, and configuring training settings to create a custom LoRA model.

Step-by-step

Go to Train a LoRA concept on Civitai
Enter a name for the model and acknowledge the terms
Upload a zip file containing your training dataset
Navigate to the Caption section and select Auto Label
Choose the base model you want to train on
Set a test prompt to generate images for each epoch
Keep the default settings and click Submit
Download the finished model and place it in your LoRA subdirectory under models

Options

Use Flux or Z image as the base model
Add a trigger word (optional)

Watch out for

All images in the dataset must be labeled so the model knows what it is looking at
Ensure all images are captioned if using Flux or Z image

Tips

Use a vision model for auto-labeling to handle the captioning process
Keep all the default training settings for a standard setup

Highlights

“Training your own custom Laura is much less intimidating than you think.”

All demos from “Local "AI Rendering" Using 3D - Explained by a Human”

3:041:18Enhance 3D clay models with Flux and ControlNetThe creator demonstrates using the Flux model in ComfyUI with a Depth ControlNet to transform a gray 3D massing model into a detailed streetcar scene while maintaining structural integrity.ComfyUI· Image to Image
6:401:29Train a custom LoRA on CivitaiCurrentA step-by-step walkthrough of uploading a dataset of streetcar images to Civitai, using auto-labeling, and configuring training settings to create a custom LoRA model.Civitai· AI Image Generator
8:512:02Image-to-image enhancement with denoisingThe video shows how to use the 'denoising' strength in ComfyUI to blend a base 3D rendering with AI-generated details like autumn trees and realistic people.ComfyUI· Image to Image
12:211:24Text-to-image generation with Z-Turbo and QwenDemonstration of the Z-Turbo model and Qwen text encoder to generate specific advertising text on the side of a 3D tram model within a ComfyUI workflow.ComfyUI· Text to Image
15:063:12Transform 3D scenes with Qwen2-VL (Qwen-Edit)The creator uses the Qwen-Edit workflow to perform complex scene modifications, such as changing weather to rain or snow, while preserving the original 3D geometry and text.ComfyUI· AI Inpainting
20:221:07Fix text and faces using Crop and StitchA demonstration of a custom 'crop image' node to isolate specific areas like signs or faces, regenerate them at native resolution, and stitch them back into the high-res image.ComfyUI· AI Inpainting
23:511:57Animate 3D renders with Wan 2.1The creator walks through a Wan 2.1 video generation workflow in ComfyUI, using the Painterly I2V node to control motion speed and animate a static 3D render.ComfyUI· Image to Video
Watch “Local "AI Rendering" Using 3D - Explained by a Human” →

AI Image Generator

6:401:29Train a custom LoRA on CivitaiCurrentA step-by-step walkthrough of uploading a dataset of streetcar images to Civitai, using auto-labeling, and configuring training settings to create a custom LoRA model.Matt Hallett Visual