What does a LoRA being
The article discusses the concept of
Why it matters
Understanding the concept of
Key Points
- 1The author is experimenting with character LoRA training for z-image-turbo using AI-Toolkit, trying to get more consistent character likeness.
- 2The author has created a workflow in ComfyUI to generate multiple versions of an image with fixed seed and steps, but different LoRAs, to see the effect side by side.
- 3The author is considering training past 3000 steps, up to 4000, and is wondering what it means for a LoRA to
- 4 in the context of character training.
Details
The author has been doing character LoRA training for z-image-turbo using AI-Toolkit, experimenting with different settings and dataset sizes to achieve more consistent character likeness. They have created a workflow in ComfyUI to generate multiple versions of an image with fixed seed and steps, but different LoRAs, to see the effect side by side. The author's latest dataset is 33 images, and they used mostly default/recommended settings from Ostris' tutorials. The likeness is pretty consistent, with the 3000 steps version usually being better than the 2750 and 2500 versions. The author is now considering training past 3000 steps, up to 4000, and is wondering what it means for a LoRA to
No comments yet
Be the first to comment