Local models & LoRAs
Run AI on your own computer for free, and teach it your style.
You need: A graphics card with at least 8 GB of memory, or an Apple Silicon Mac.
Watch it in the app
Watch the full tutorial on YouTube ›
Get a model
- Open Settings ▸ Local models.
- Find a model. The table below shows how big each one is.
- Press download. Then pick it like any other model.
Add a LoRA
- Pick a ComfyUI model and open its Settings.
- Press + Add a LoRA. A LoRA is a small file that teaches a model a style or a character.
- Set its strength. Add to prompt puts its special words in for you.
This works on every image and video model on ComfyUI, on this computer or another one. It does not work on Comfy Cloud. A model can hold up to four LoRAs.
Settings on every model
Seed, steps, guidance, sampler, scheduler and denoise. Not sure? Press Suggest settings. Use these settings on a finished result puts its settings back.
How big is each model?
One click each in Settings ▸ Local models. Sizes are a fresh download: models share some files, so two often take less than the sum. Object tools and Wan VACE Reframe are far smaller with the quick 1.3B fill only (about 13 and 11 GB).
| Model | Windows · Linux | Mac |
|---|---|---|
| Video | ||
| MiniMax H3 | 44 GB | not on Mac |
| MiniMax H3 Ref · Omni Ref | 44 GB | not on Mac |
| LTX 2.5 | 47 GB | 66 GB |
| LTX 2.3 | 23 GB | 73 GB |
| Wan 2.2 14BNEW | 36 GB | 66 GB |
| Wan 2.2 5B | 18 GB | 22 GB |
| Wan 2.2 Fun Control HQ | 34 GB | 68 GB |
| Wan Fun Control | 11 GB | 15 GB |
| Image | ||
| FLUX.2 dev | 53 GB | 97 GB |
| Qwen-Image Edit | 30 GB | 55 GB |
| Qwen Edit Plus 2511 | 30 GB | not on Mac |
| Qwen-Image 2512 | 30 GB | 57 GB |
| Qwen-Image | 29 GB | 54 GB |
| Z-Image Turbo | 20 GB | 20 GB |
| FLUX.2 Klein 9B + Edit | 18 GB | 35 GB |
| Qwen-Image 2.1 + Edit | 17 GB | not on Mac |
| FLUX.1 dev | 17 GB | not on Mac |
| SDXL + Edit | 6.5 GB | 6.5 GB |
| Edit & finish | ||
| Object tools (SAM 3.1 + Wan VACE) | 45 GB | 49 GB |
| Wan VACE Reframe | 43 GB | 47 GB |
| Object · VOID removerNEW | 32 GB | 32 GB |
| Object · fast clean plateNEW | 1.1 GB | 1.1 GB |
| 3D | ||
| Hunyuan3D 2.0 | 4.9 GB | 4.9 GB |
| Sound | ||
| ACE-Step 1.5 XL turbo | 18 GB | not on Mac |
| MiniMax Music 3 | 11 GB | not on Mac |
| Stable Audio 3 Medium | 9.7 GB | not on Mac |
| ACE-Step 1.5 turbo | 9.3 GB | 9.3 GB |
| Stable Audio 3 SFX | 3.2 GB | not on Mac |
| Stable Audio 3 Music | 3.2 GB | not on Mac |
The assistant’s brain on your computer
Through Ollama, free.
| Model | Download | For |
|---|---|---|
| Gemma 4 12B | 8 GB | 16 GB card |
| Gemma 4 E4B | 6.6 GB | 8 to 12 GB card |
| Gemma 4 E2B | 4.6 GB | laptops |
Stuck? Press the round button at the bottom right of the app and ask the assistant. It knows every screen. Or ask other editors in The VFX Room.