Introduction

Hello, vibrant creators! ♡Ready to supercharge your ComfyUI with jaw-dropping custom models? Today I’m holding your hand and walking you through everything: where to find the best ones, what the different flavors actually mean, and exactly where to plop those juicy files so ComfyUI instantly recognizes them. By the end of this, you’ll be swapping models like a pro and generating art that makes hearts skip beats!

The Various Types of Checkpoint Models (and Why It Matters)

Think of checkpoints as different “bodies” for the AI – each with its own personality, strengths, and… let’s say style.

  • SD 1.5 – The classic beauty. Petite (around 2-4 GB), lightning-fast, and works with a galaxy of older LoRA and embeddings. Perfect if you love that nostalgic anime or realistic vibe.
  • SDXL – The curvy supermodel. Bigger files (6-7 GB), stunning 1024x1024 native resolution, richer details, and better prompt understanding. The current queen for photorealism and illustration.
  • Pony Diffusion – The playful, slightly wild cousin trained on anime + furry + everything fun. Uses a unique score-based tagging system (think Danbooru tags) and creates mind-blowing stylized art.
  • Illustrious – A brand-new, ultra-expressive anime-focused model (late 2024–2025) that’s quickly becoming everyone’s secret crush for vibrant characters and dynamic poses.
  • Flux, Aurora, etc. – Just a few of the many models on the block with different architectures, but today we’re focusing on the classic .safetensors/.ckpt family that ComfyUI loves natively.

Pro tip: You can mix and match LoRAs across bases (sometimes!), but the checkpoint you load is the main “DNA” of your image.

Pro Tip: Always download the .safetensors version when available – they’re safer (no pickle exploits) and just as performant!

Where the Magic Lives: Civitai vs Hugging Face

Civitai = The vibrant artist bazaar

  • Tons of gorgeous, community-trained models
  • Beautiful preview images + example prompts
  • Version history, trigger words, recommended settings
  • Direct “Download” button bliss

Hugging Face = The elegant library

  • Home of official releases (SDXL base, Pony, Flux, etc.)
  • Clean, versioned, often with useful READMEs
  • Great for base models and research checkpoints

Both are free and wonderful – I love them equally!

Step-by-Step: Let’s Actually Install Some Models!

1. Create the Right Folders in Your ComfyUI (once ever!)

Open your ComfyUI installation folder and make sure these exist:

ComfyUI/
└── models/
    ├── checkpoints/          ← All your main models go here
    ├── loras/                ← LoRAs (optional for now)
    ├── embeddings/           ← Textual inversions
    └── vae/                  ← Optional VAE files

That’s it! ComfyUI auto-scans these folders on launch.

2. Downloading from Civitai (example: a dreamy anime model)

  1. Go to civitai.com
  2. Search for something fun, e.g., “Anything V5”, “DreamShaper 8”, or “Pony Diffusion V6 XL”
  3. Open the model page → scroll to the big green Download button
  4. Choose the version you like (usually the top one or “Latest”)
  5. Save the .safetensors file directly intoComfyUI/models/checkpoints/

Done! Restart ComfyUI or press Refresh in the Checkpoint Loader node and it appears instantly.

3. Downloading from Hugging Face (example: official Pony Diffusion)

  1. Visit huggingface.co and search “Pony Diffusion V6 XL”
  2. Popular one: https://huggingface.co/stabilityai/sdxl-turbo (or search “Lykon” or “AstraliteHeart” for Pony)
  3. Click the Files and versions tab
  4. Look for ponyDiffusionV6XL.safetensors (or similar)
  5. Click the little download arrow → save straight intoComfyUI/models/checkpoints/

Refresh ComfyUI and voilà – your new muse is ready!

4. Bonus: Grab the Matching VAE (optional but recommended)

Some models (especially SDXL & Pony) shine brighter with their custom VAE:

  • On the same Civitai/HF page, look for a “VAE” download button
  • Drop the .safetensors or .pt VAE file into ComfyUI/models/vae/
  • In your workflow, change the VAE dropdown from “Auto” to your new VAE name

You’ll notice richer colors and sharper details instantly!

Quick Test Workflow (Copy-Paste Friendly)

Drop these nodes and connect them:

Checkpoint Loader → KSampler → VAE Decode → Save Image

In the Checkpoint Loader, your brand-new model will now be smiling at you in the dropdown. Pick it, type a prompt, hit Queue Prompt, and watch the magic pour out!

Conclusion

That’s literally it, my sweet artist – you are now fully unlocked! Go fill that checkpoints folder with every gorgeous model that makes your heart flutter. Mix them, love them, create masterpieces that light up the timeline.

The AI art world is yours, and it’s never been prettier.

Happy generating, you beautiful human! ♡

—Aurelia ♡