I will train a custom multi model lora

People, original characters, illustrated characters, products, objects, mascots, visual styles, fashion concepts and other suitable subjects.

No. Training material can be photographic, illustrated, stylized or 3D depending on the project and target architecture.

Yes. The LoRA can learn identity characteristics while compatible models and prompting can generate different visual treatments.

No. LoRAs are architecture-specific. If you need the concept available across multiple model families, dedicated versions need to be trained.

FLUX and Stable Diffusion/SDXL are primary target families. Other architectures can be evaluated before ordering.

Yes. I can recommend the target model according to the subject, intended use, desired fidelity, workflow, hardware and licensing requirements.

There is no universal number. Quality, diversity and consistency matter more than simply adding more images. Send your current dataset and I will evaluate it.

Yes, provided the selected architecture and licensing conditions permit delivery of the trained weights.

A revision covers reasonable tuning within the original training objective. A completely different subject, dataset or target model requires a new training scope.

That depends on your rights to the training material and the license of the selected base model. Commercial requirements should be stated before training.