Civitai models
Civitai Models Explained: Checkpoints, LoRAs, Versions and Downloads
Civitai uses “model” as an umbrella term for several kinds of resources. Choosing the correct type and version prevents most failed downloads, broken workflows, and disappointing generations.
- 01 Type Checkpoint or add-on
- 02 Base Match the architecture
- 03 Version Use exact files
Civitai model types at a glance
A checkpoint is a complete model that can generate on its own. A LoRA is a smaller adapter applied to a compatible checkpoint. Textual inversions add learned tokens, while ControlNet-style resources guide structure such as pose, depth, or edges. Workflows describe the graph and settings used by tools such as ComfyUI.
The category tells you how to install and invoke the resource. It does not tell you whether it matches your current base model, so always check both fields.
- Checkpoint: full generation model, usually the largest download.
- LoRA or LyCORIS: compact add-on used at an adjustable weight.
- Embedding or textual inversion: learned token file used in a prompt.
- Control resource: structural guidance for a compatible pipeline.
- Workflow: nodes, connections, and settings rather than model weights.
Base-model compatibility comes first
A resource trained for one architecture usually cannot be moved blindly to another. Match labels such as SD 1.5, SDXL, Pony, Illustrious, Flux, or another named base to the checkpoint and runtime you plan to use.
Derivative checkpoints can sometimes share LoRAs with their base family, but behavior varies. Start with the creator’s recommended base, trigger words, and weight, then change one variable at a time.
Choose the right version and file
A Civitai model can have many versions. Each version may change the base model, trained words, files, and expected output. Select the version used by the example you want to reproduce instead of automatically downloading the newest one.
Prefer documented formats supported by your runtime. Review file size, hashes, and Civitai scan results. A hash lets tools identify an exact file even if someone renamed it.
Note: A successful platform scan reduces risk but does not replace normal caution with downloaded files and custom workflow nodes.
Evaluate quality, permissions, and provenance
Example quality depends on prompts, cherry-picking, post-processing, and linked resources. Look for several examples from different users, visible generation metadata, clear version notes, and creator responses to questions.
Read usage permissions before commercial work, redistribution, hosted services, or derivative training. Save the model URL and version ID with your workflow so you can trace what you used later.
Questions
Civitai FAQ
What is the difference between a Civitai checkpoint and LoRA?
A checkpoint is a complete generation model. A LoRA is a smaller adapter that changes a compatible checkpoint.
What does base model mean on Civitai?
It identifies the architecture or model family used to train the resource. Your checkpoint and runtime need compatible support.
Why are there multiple files in one version?
Creators may publish different precision, format, training-epoch, or component variants. Read the file labels and version notes before choosing.
Connect Civitai to Klopaper
Add your own API key for reliable discovery, gated LoRAs, and Civitai collections. Your key is encrypted and isolated to this browser session.
Primary sources
Klopaper is independent and is not affiliated with Civitai. Product interfaces, prices, and policies can change.