Learn Civitai
The community hub for Stable Diffusion and Flux models, LoRAs and embeddings, with model cards, example settings and an on-site generator.
What Civitai is good at
- Finding checkpoints and LoRAs with example images and settings
- Trigger words and recommended weights on every model card
- On-site generation without a GPU
- Licences set per model
Quickstart
- Read the model cardBase model, trigger words, recommended sampler, steps and weights are all there.
- Open an example imageIts generation data shows the exact prompt and settings; start from those.
- Add a LoRA at low weight0.6 to 0.8 first; stack a second one only after the first behaves.
- Check the licencePermissions for commercial use and redistribution are shown per model; keep a note.
Settings cheat-sheet
| Setting | What it does | Start with |
|---|---|---|
| Checkpoint | Base model | The one the LoRA was trained on |
| LoRA weight | Strength of the add-on | 0.6 to 0.8 |
| Trigger words | From the model card | Include them |
| Sampler / steps | From the example data | Copy, then adjust |
| CFG | Prompt adherence | 5 to 7 |
| Licence | Per model | Check before client work |
Civitai guides
Guides for Civitai are on their way. The quickstart and cheat-sheet above are kept current in the meantime.
The five most-asked questions
What is Civitai?
A community site where people publish Stable Diffusion and Flux checkpoints, LoRAs, embeddings and workflows, with example images that carry their generation settings. It also has an on-site generator.
What is a LoRA and how do I use it?
A LoRA is a small add-on that teaches a model a style, character or object. Load it with the matching base model, use its trigger words and apply it with a weight, usually below 1.
Can I generate on Civitai without my own GPU?
Yes, the on-site generator runs supported models and LoRAs with the usual settings and is paid in Buzz, the site's credit.
Are Civitai models safe to use commercially?
It depends: every model shows permissions set by its uploader and the base model's licence applies too. Read both and record which models made each image.
How do I reproduce an example image?
Open the image, copy its generation data (prompt, negative, seed, sampler, steps, CFG, size, model hash) and run it with the same checkpoint and LoRAs.