neptunite/moodyDesireMix_GGUF
neptunite • imageMoody Desire Mix v1.0 - GGUF Quantizations
GGUF quantized versions of Moody Desire Mix v1.0 (Flux.2 Klein 9B Checkpoint), quantized from the original FP16 source using the city96/ComfyUI-GGUF toolchain.
GGUF 量化版本，原始模型来自 Moody Desire Mix v1.0 （Flux.2 Klein 9B Checkpoint），使用 city96/ComfyUI-GGUF 工具链从 FP16 源文件量化而来。
Available Versions / 可用版本
| File / 文件 | Quant / 量化 | Size / 大小 | VRAM / 推荐显存 | Quality / 画质保留 |
|---|---|---|---|---|
moodyDesireMix_v10_Q8_0.gguf | Q8_0 | ~10 GB | 12GB+ | 98–99% |
moodyDesireMix_v10_Q6_K.gguf | Q6_K | ~8 GB | 10GB+ | 95–97% |
moodyDesireMix_v10_Q5_K_M.gguf | Q5_K_M | ~7 GB | 8GB+ | 90–93% |
moodyDesireMix_v10_Q4_K_M.gguf | Q4_K_M | ~6 GB | 6GB+ | 80–87% |
Recommendation: Use Q8_0 if VRAM allows. Q6_K / Q5_K_M for 6–10GB cards. Q4_K_M for 4–6GB cards as a last resort.
建议：显存充裕优先选 Q8_0；6–10GB 显卡推荐 Q6_K 或 Q5_K_M；4–6GB 显卡选 Q4_K_M。
Usage (ComfyUI) / 使用方法（ComfyUI）
Step 1 — Install ComfyUI-GGUF / 安装 ComfyUI-GGUF
Search for ComfyUI-GGUF in ComfyUI Manager and install it, or clone manually:
在 ComfyUI Manager 中搜索并安装 ComfyUI-GGUF，或手动克隆：
cd ComfyUI/custom_nodes
git clone https://github.com/city96/ComfyUI-GGUF
Step 2 — Place the Model File / 放置模型文件
Place the downloaded .gguf file into:
将下载的 .gguf 文件放入：
ComfyUI/models/unet/
Step 3 — Load in Workflow / 在 Workflow 中加载
In the ComfyUI canvas:
- Double-click to open the node search
- Search for
Unet Loader (GGUF)and replace your existing Load Diffusion Model node - Select the
.gguffile from the dropdown
在 ComfyUI 画布中：
- 双击空白处打开节点搜索
- 搜索
Unet Loader (GGUF)，替换原有的 Load Diffusion Model 节点 - 在节点下拉菜单中选择对应的
.gguf文件
About GGUF Quantization / 关于 GGUF 量化
GGUF is a quantization format popularized by llama.cpp that supports offloading model layers to system RAM, enabling large models to run even when VRAM is insufficient. Q8_0 has negligible quality loss compared to FP16, while Q5_K_M offers the best balance between file size and image quality.
GGUF 是 llama.cpp 推广的量化格式，支持将模型层卸载到系统内存（RAM）， 在显存不足时也能运行大模型。Q8_0 与 FP16 原版质量差距极小，Q5_K_M 是文件大小与画质的最佳折中选择。
Original Model / 原始模型
- Original / 原始模型：Moody Desire Mix v1.0 on CivitAI
- Base Architecture / 基础架构：Flux.2 Klein 9B
- Quantization Tool / 量化工具：city96/ComfyUI-GGUF tools