The following are current known limitations which we plan to address in future updates:
General limitation
You can only keep one style-based and one product-based LoRA per workspace at a time. To train a new LoRA, existing models must be deleted first. This limitation will be removed in future updates. It takes around 3-4 hours to train a new LoRA with Bria.
Known Product Fidelity challenges
Product LoRA outcomes are not as advanced as those of Style or Object based LoRAs. Following the below guidelines will maximise the likelihood of your desired outcome:
Low Fidelity (Generic products) → Easier to train, slight randomness allowed.
Medium Fidelity (Semi-detailed products) → More structured training needed.
High Fidelity (Brand-critical products) → Difficult to perfect, logos and brand text must be clear in training images.
We are working to support LoRAs for other base models - keep your eyes out for future updates.