Diffusion Docs

Appearance

Hosted Or Local? ​

Hosted and local generation are both useful. The right choice depends on what you are optimizing for.

Quick Take

Hosted tools reduce setup friction. Local tools increase control. Many users start hosted, then move local when they need custom models, LoRAs, or repeatable workflows.

Hosted Generation ​

Hosted generation runs on someone else's servers. You open a website, enter a prompt, and generate without configuring Python, drivers, model folders, or GPU settings.

Good fits:

  • fast experimentation
  • low-power laptops or phones
  • conversational editing
  • design platforms that include generation

Common limits:

  • model choices may be hidden or restricted
  • settings may be simplified
  • terms, privacy, and commercial-use rules can vary
  • exact reproducibility may be limited

Local Generation ​

Local generation runs on your own machine or a rented GPU. It usually gives more access to models, settings, LoRAs, extensions, metadata, and workflow files.

Good fits:

  • custom checkpoints and LoRAs
  • repeatable workflows
  • private exploratory images or client work
  • advanced inpainting, image-to-image, and control workflows

Common limits:

  • setup can be fragile
  • hardware matters
  • updates can break extensions or environments
  • new model families may require UI/backend updates

Hybrid Is Normal

You do not have to pick one forever. It is common to use hosted tools for ideation and local tools for controlled workflows, LoRAs, private work, or batch generation.