Civitai

Civitai is an open image model community run by a US company, and the de facto central repository of the Stable Diffusion ecosystem.

Nearly every popular checkpoint, LoRA and embedding lives here. Each model page shows sample images uploaded by the author, many with full generation parameters attached for reproduction. Beyond downloads it also offers online generation and training, paid for with on-site credits.

If you run SD or FLUX locally, you will end up here: finding models, seeing how others use them, and downloading them all happen in one place. The community is lively too, with tutorials, bounties for requested models and plenty of discussion.

Key features

• Model hosting and downloads: Checkpoints, LoRAs, embeddings, ControlNet files and more, filterable by base model, type and popularity.
• Samples with parameters: Example images carry the prompt, sampler, steps and so on, which makes reproduction far faster than guessing.
• Online generation: Generate with hosted models in the browser without downloading anything, paid in credits.
• Online training: Upload images and train a LoRA in the cloud — useful without a GPU.
• Community content: Articles, model bounties and galleries for learning and sharing.

How to use

1. Sign up at civitai.com and set your content filter level in the settings first.
2. Filter models by base model version and make sure any LoRA matches the checkpoint you use.
3. Review the samples, recommended settings and licence on the model page, then download into the right local folder.
4. If you would rather not run locally, generate or train online with credits.

Best for

• Anyone running SD or FLUX locally: the first place to find and download models.
• Reproducing a specific look: copy the sample’s parameters instead of trial and error.
• Niche styles: the catalogue is large enough that someone has often already made one.
• Learning to train models: plenty of public LoRAs and write-ups to learn from.

Strengths and limitations

Strengths

• The largest and fastest-moving catalogue among comparable communities.
• Samples with parameters make results easy to reproduce.
• Browsing and downloading are free.

Limitations

• Moderation is relatively permissive, so adjust the filters before browsing; the platform’s content policies also change from time to time.
• Licences vary by model and some forbid commercial use; training-data provenance and likeness rights are yours to assess.
• Downloads from mainland China can be slow, especially for large model files.
• The interface is dense and can overwhelm newcomers.