RunComfy: Reviewing LoRA training sample results
RunComfyDemo summary
The creator reviews the output samples from the training process, comparing different step intervals to identify the best art style replication for the D&D monster LoRA.
Step-by-step
- Review the output samples generated during the LoRA training process
- Compare the visual results across different step intervals
- Identify the point where the art style most closely matches the target D&D aesthetic
- Check for signs of overtraining in the later step samples
Watch out for
- Training up to 4,000 steps may result in overtraining for this specific model
Tips
- Aim for approximately 3,000 steps to achieve the best balance of style replication without overtraining
Highlights
“consistently as it goes down, it gets closer to the art style that we're trying to achieve”
All demos from “Train a LoRA for LTX 2.3 / Z-Image on RunComfy | Dataset Captioning + AI Toolkit Tutorial”
- 6:381:58Creating a new dataset in RunComfy TrainerThe user shows how to upload a zipped folder of images and matching text captions to the RunComfy Trainer interface to define a new training dataset.RunComfy· AI Avatar Generator
- 8:362:41Configuring a LoRA training job on RunComfyThe demonstration covers setting up a training job for the Z-Image architecture, including setting the trigger word, step count, and sample prompts for monitoring progress.RunComfy· AI Avatar Generator
- 13:251:10Reviewing LoRA training sample resultsCurrentThe creator reviews the output samples from the training process, comparing different step intervals to identify the best art style replication for the D&D monster LoRA.RunComfy· AI Avatar Generator
- Watch “Train a LoRA for LTX 2.3 / Z-Image on RunComfy | Dataset Captioning + AI Toolkit Tutorial” →
AI Avatar Generator
- 6:381:58Creating a new dataset in RunComfy TrainerThe user shows how to upload a zipped folder of images and matching text captions to the RunComfy Trainer interface to define a new training dataset.Eye For AI
- 8:362:41Configuring a LoRA training job on RunComfyThe demonstration covers setting up a training job for the Z-Image architecture, including setting the trigger word, step count, and sample prompts for monitoring progress.Eye For AI
- 13:251:10Reviewing LoRA training sample resultsCurrentThe creator reviews the output samples from the training process, comparing different step intervals to identify the best art style replication for the D&D monster LoRA.Eye For AI