Choose A Generation Path
Before choosing settings, choose the kind of generation path you want. Diffusion tools range from simple hosted websites to local interfaces with custom models and advanced workflow control.
Quick Take
Use hosted generation to learn prompting quickly. Use local generation when you need custom models, LoRAs, privacy, repeatability, or advanced control.
The Main Paths
| Path | Best For | Tradeoff |
|---|---|---|
| Hosted website | Fast start, low setup, conversational editing | Less control over models, settings, privacy, and reproducibility |
| Local UI | Custom checkpoints, LoRAs, inpainting, repeatable workflows | Requires hardware, installation, updates, and troubleshooting |
| Manager or launcher | Trying several local UIs and sharing model folders | Adds another layer to understand |
| Developer library | Automation, apps, experiments, reproducible pipelines | Requires coding and more technical setup |
For a broader comparison, see Generation Tools.
Start Simple
If you are learning, start with one path and one workflow. Switching tools too early can make it hard to tell whether a result changed because of the model, prompt, sampler, scheduler, UI defaults, or hidden backend behavior.
Do Not Compare Tools By One Image
A single seed and prompt can make one tool look better by chance. Compare tools by workflow control, model support, output consistency, maintenance, and how easy it is to diagnose failures.
A Practical Decision
Choose hosted generation if you want to:
- learn prompt iteration
- make quick exploratory images
- avoid installing dependencies
- work from low-power hardware
Choose local generation if you want to:
- use custom checkpoints, VAEs, LoRAs, or embeddings
- keep prompts and outputs on your machine
- save exact settings and seeds
- use inpainting, ControlNet-style inputs, regional prompting, or custom nodes