README.md

---
license: mit
tags:
- comfyui
- models
- gguf
- safetensors
- flux
- wan
- qwen
- hidream
- ltxv
---

# ComfyUI Auto-Installer — Assets

[![HuggingFace](https://img.shields.io/badge/🤗_HuggingFace-UmeAiRT-yellow.svg)](https://huggingface.co/UmeAiRT/ComfyUI-Auto_installer)
[![ModelScope](https://img.shields.io/badge/🔬_ModelScope-UmeAiRT-blue.svg)](https://www.modelscope.ai/datasets/UmeAiRT/ComfyUI-Auto-Installer-Assets)
[![License](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)

Pre-packaged model files and binaries for the UmeAiRT ecosystem. This repository is a **pure asset store** — no code, only large binary files served via Git LFS.

**Consumed by:**
- [ComfyUI-Auto_installer](https://gitlab.com/UmeAiRT-Studio/ComfyUI-Auto_installer-Python) — automated Windows setup, downloads models from this repo
- [ComfyUI-UmeAiRT-Toolkit](https://gitlab.com/UmeAiRT-Studio/ComfyUI-UmeAiRT-Toolkit) — custom ComfyUI nodes with automatic model downloading

## 📦 What's Included

### Diffusion Models (`models/diffusion_models/`)

Each model family provides multiple precision levels: full precision (bf16/fp16), reduced (fp8), and GGUF quantizations (Q8→Q3) for different VRAM budgets.

| Family | Models | Size | Precisions |
|---|---|---|---|
| **FLUX** (12B) | Dev, Fill, Kontext, Canny, Depth, Schnell | 384 GB | fp16, fp8, SVDQ-int4, Q8→Q3 |
| **WAN 2.1** (14B) | T2V, I2V-480p/720p, VACE, Fun-InP/Control | 1575 GB | bf16, fp16, fp8, Q8→Q3 |
| **WAN 2.2** (A14B) | T2V, I2V, InP, Control, Camera (High/Low) | *(incl. above)* | fp16, fp8_scaled, Q8→Q3 |
| **QWEN** (~20B) | Image, Image-Edit, Image-Distill | 412 GB | bf16, fp8, SVDQ, Q8→Q3 |
| **HiDream** (17B) | I1-Dev | 76 GB | fp8, Q8→Q3 |
| **LTX-2** (19B+22B) | 19B-Dev, 2.3-22B-Dev | 140 GB | Q8→Q3 |
| **LTXV** (13B) | 0.9.8-Dev | 46 GB | Q8→Q3 |
| **Z-Image** (12B) | Standard, Turbo | 85 GB | bf16, fp8, Q8→Q3 |

### Text Encoders (`models/text_encoders/`)

| Encoder | Precisions |
|---|---|
| **T5 XXL** | fp16, fp8, fp8_scaled, Q8→Q3 |
| **umt5 XXL** | bf16, fp8_scaled, Q8→Q3 |
| **GEMMA-3 12B** | bf16, fp8, IQ4_XS, Q8→Q3 |
| **Qwen2.5-VL-7B** | bf16, fp8_scaled, UD-Q4, Q8→Q3 |
| **Qwen3-4B** | bf16, fp8_mixed, UD-Q8→Q3 |
| **LLAMA 3.1 8B** | fp8_scaled |

### Other Models

| Folder | Contents |
|---|---|
| `models/clip/` | CLIP-L, CLIP-G, LongCLIP, ViT-L-14 |
| `models/clip_vision/` | clip_vision_h, SigCLIP |
| `models/checkpoints/` | Full merged checkpoints |
| `models/loras/` | LoRA adapters (FLUX, WAN, QWEN, LTX-2) |
| `models/vae/` | Shared VAE models |
| `models/controlnet/` | ControlNet models |
| `models/upscale_models/` | RealESRGAN and others |
| `models/style_models/` | Style/Redux models |
| `models/sams/` | SAM segmentation |
| `models/ultralytics/` | YOLO face/hand detection |
| `models/LLM/` | Florence-2 PromptGen |

### Binaries & Packages

| Folder | Contents |
|---|---|
| `bin/` | Git, Python, CUDA installers |
| `whl/` | Pre-compiled Python wheels |
| `packages/` | Extension zips, build tools, configs |

## 📐 Naming Conventions

**GGUF** — CamelCase per [GGUF spec](https://github.com/ggml-org/ggml/blob/master/docs/gguf.md):
```
Flux1-Dev-Q4_K_S.gguf
Wan2.2-T2V-HighNoise-14B-Q8_0.gguf
```

**Safetensors** — lowercase with hyphens:
```
wan2.1-t2v-14b-fp16.safetensors
qwen-image-edit-fp8_e4m3fn.safetensors
```

## 🔗 Related Projects

| Project | Description |
|---|---|
| [ComfyUI-Auto_installer](https://gitlab.com/UmeAiRT-Studio/ComfyUI-Auto_installer-Python) | PowerShell installer that downloads from this repo |
| [UmeAiRT-ComfyUI-Toolkit](https://gitlab.com/UmeAiRT-Studio/ComfyUI-UmeAiRT-Toolkit) | Custom ComfyUI nodes |

## 📜 License

MIT — see [LICENSE](LICENSE) for details.

![UmeAiRT](images/UmeAiRT.png)