LoRA & Training Blocks
Use training blocks to fine-tune image generation with LoRA weights and reuse them across bleuprints.
Available blocks:
- LoRA Image Finetune: trains a LoRA model from a set of images and a trigger word.
- Save Train Object: stores trained weights with a label and description for later reuse.
- Load Fine Tune: loads a saved LoRA (or other train types as supported).
- Fine-tuned Image Generation: generates images using a selected LoRA model with standard controls.
How to use (typical flow):
- Provide a batch of training images to LoRA Image Finetune.
- Configure a trigger word and training steps.
- Save the resulting train object using Save Train Object.
- In another workflow, Load Fine Tune and connect it to Fine-tuned Image Generation along with your prompt.
Known constraints:
- Training expects a list of images.
- Train types must match when connecting training outputs to inputs. Some inputs accept any, while others require a specific type (e.g., Flux LoRA).
- Fine-tuned generation supports standard image options like aspect ratio, guidance, and seed.