How to Download Civitai Models — Step-by-Step Guide

What You Need

• A CivitAI account (free, optional but recommended for NSFW models)
• A Stable Diffusion UI installed — Forge, ComfyUI, or A1111
• 2-25 GB disk space per model depending on architecture
• Internet connection (for downloading — running is fully offline)

Step 1 — Create a CivitAI Account

Go to civitai.com and sign up. It is free.

Some NSFW models require login to see and download. The NSFW content toggle is off by default — enable it in your account settings under Content Preferences to see all models.

You can browse and download most SFW models without an account. But the account also lets you save favorites, track downloads, and leave reviews.

Step 2 — Find a Model

Use the search bar or browse by category. The filters that matter most:

• Model Type: Checkpoint, LoRA, Textual Inversion, VAE
• Base Model: SDXL, Flux, SD 1.5 — pick the one matching your installed setup
• Sort: "Most Downloaded" or "Highest Rated" to find proven models

Do not download an SDXL model if you only have an SD 1.5 setup — it will not load. Check the "Base Model" tag on every model page.

Not sure which models to start with? See our Civitai Models Download List for curated picks.

Step 3 — Choose the Right Version

Most models have multiple versions in a dropdown on the model page. General rules:

• Pick the latest version unless reviews say the older one was better
• Look for "BakedVAE" variants — they include the VAE so you do not need a separate download
• Check the description for version-specific changes (hand fixes, style changes, new training data)

Step 4 — Download the File

Click the blue Download button. If given a choice of file format:

• .safetensors — always pick this. Faster to load, safer (no hidden code risk)
• .ckpt — older format, carries a small security risk. Only use if .safetensors is not available
• .pt — used for textual inversions/embeddings only

Some models also offer "pruned" vs "full" versions. Pruned files are smaller but lose some quality. For most users, pruned is fine.

Step 5 — Place the File in the Correct Folder

Forge / A1111:

ComfyUI:

Just drag the downloaded file into the matching folder.

Step 6 — Load the Model in Your UI

• Forge / A1111: Click the refresh button next to the checkpoint dropdown at the top of the page. Your new model appears in the list. Select it.
• ComfyUI: The model shows up automatically in the Load Checkpoint node dropdown. If it does not, restart the UI.

Verify It Works

Select your model and type a simple test prompt: "portrait of a woman, photorealistic" for realistic models, or "score_9, score_8_up, 1girl, anime style" for Pony/anime models. Click Generate. If you get an image, you are done.

Troubleshooting

• Model does not appear in dropdown: Wrong folder, or you forgot to click refresh. Double-check the folder path.
• "Out of memory" error on load: Model is too large for your VRAM. Try a pruned/FP16 variant, or switch to a smaller architecture (SD 1.5 if you are trying SDXL).
• Black or corrupt images: VAE not loaded. Download the matching VAE file and place it in the VAE folder, or use a BakedVAE version of the model.
• Loads but produces garbage: Wrong base model — e.g., SDXL model loaded in an SD 1.5 pipeline. Check the model base model tag on CivitAI.
• Slow download speeds: CivitAI gets congested during peak hours (US evenings). Try early morning or use a download manager.

What to Do Next

• Run your first generation: How to Run Civitai Models Locally — full setup from UI install to first image
• Browse top models: Civitai Models Download List — curated picks across all architectures
• Compare realistic options: Best Realistic Civitai Models — Juggernaut, RealVisXL, epiCRealism ranked

Or skip the manual setup entirely — LocalForge AI comes with Forge pre-configured and popular models already installed.