CivArchive
    Mossimage-BF16-Enhanced Version - 1.0
    Preview 1
    Preview 2
    Preview 3
    Preview 4
    Preview 5
    Preview 6
    Preview 7
    Preview 8
    Preview 9
    Preview 10
    Preview 11

    Description

    MOSS Image 是由 OpenMOSS 团队开发的一款 5B 参数文本对图像 DiT,旨在实现准确的图像内文本、布局控制和长提示。它将Qwen3.5 9B文本编码器与FLUX.2 VAE配对,接受最多4K的自然语言标记、booru风格标签或JSON布局的提示,并能在任何宽高比下渲染512px至2K。

    📂 ComfyUI/

    ├── 📂 models/

    │ ├── 📂 diffusion_models/

    │ │ ├── moss_image_base_bf16.safetensors

    │ │ ├── moss_image_base_int8_convrot.safetensors

    │ │ ├── moss_image_sft_bf16.safetensors

    │ │ └── moss_image_sft_int8_convrot.safetensors

    │ ├── 📂 text_encoders/

    │ │ └── qwen3.5_9b_w4a8_***q.safetensors

    │ │ └── qwen3.5_9b_int8_convrot.safetensors

    │ ├── 📂 vae/

    │ │ └── flux2-vae.safetensors

    根据经下参数进行设置:

    采样器:欧拉

    调度器:sgm_uniform

    步数:50

    CFG:3

    Details

    Downloads
    0
    Platform
    ShakkerAI
    Platform Status
    Available
    Created
    11/4/2026
    Updated
    11/16/2026
    Deleted
    -

    Files