CivArchive
    MOSS-Image Base & SFT — Minimal ComfyUI Workflows for Photography, Text & Layout - v1.0
    NSFW
    Preview 146531555
    Preview 146531568
    Preview 146531557
    Preview 146531558
    Preview 146531569
    Preview 146531571

     

    Click here free try online:

     

    工作流:Moss-Sft文生图-极简

    体验地址:https://www.runninghub.ai/zh-cn/post/2113166465460400130/?inviteCode=rh-v1401

     

    工作流:Moss-Base文生图

    体验地址:https://www.runninghub.ai/zh-cn/post/2113166468326535170/?inviteCode=rh-v1401

     

    工作流:Moss-Base Fixed文生图

    体验地址:https://www.runninghub.ai/zh-cn/post/2113166968816054273/?inviteCode=rh-v1401

     

    工作流:Moss Sft Fixed文生图

    体验地址:https://www.runninghub.ai/zh-cn/post/2113167471804153858/?inviteCode=rh-v1401

     

     

    Two streamlined, 11-node text-to-image workflows for photographic images, cinematic scenes, posters, thumbnails, infographics, and concept art. MOSS-Image supports Chinese and English prompts, bilingual text rendering, and experimental Layout JSON input for guiding text placement and object arrangement. The package includes the pretrained Base model and its supervised fine-tuned SFT counterpart. Model details

    Settings saved in these workflows

    Workflow Steps CFG Sampler Scheduler

    Base 25 3 er_sde simple

    SFT 25 3 euler sgm_uniform

     

     

    Official reference settings

    For the official ComfyUI plugin example, use 50 steps, CFG 3, euler, sgm_uniform, and denoise 1.0. Official example

    The official Base and SFT pipelines also use Shift 5 and negative-branch layer skipping at index 22, counted from zero. These two minimal graphs do not expose separate controls for those settings. Base settings, SFT settings

    Resolution

    Official guidance recommends approximately 1, 1.5, or 2 megapixels. The documented extrapolation range is approximately 0.25–4 megapixels, measured by total pixel count. Both dimensions must be multiples of 16. Resolution guide

    Practical size examples:

    Format Width × Height Approximate pixels

    Square 1024 × 1024 1 MP

    Portrait 1024 × 1536 1.5 MP

    Widescreen 1936 × 1088 2 MP

     

     

    Both supplied workflows currently select 16:9, approximately 2 MP. Select your local model, text encoder, and VAE files, then edit Text (User Prompt) and generate. For Layout input, paste the compiled JSON into that prompt field and match the generation dimensions to the layout canvas.

    MOSS-Image Base 与 SFT:摄影、文字与画面布局的极简 ComfyUI 工作流

    分享两套 11 节点的极简文生图工作流,分别对应 MOSS-Image Base 和 SFT,适合探索摄影质感、电影场景,以及海报、视频封面、信息图和美术概念等创作。模型支持中英文提示词、中英文文字渲染,还能通过实验性的 Layout JSON 描述文字与物体的位置。Base 为预训练版本,SFT 为其监督微调版本。 模型说明

    这两份工作流当前保存的参数

    工作流 步数 CFG 采样器 调度器

    Base 25 3 er_sde simple

    SFT 25 3 euler sgm_uniform

     

     

    官方参考参数

    官方 ComfyUI 插件示例采用:50 步、CFG 3、euler 采样器、sgm_uniform 调度器、降噪强度 1.0。 官方示例

    官方 Base 和 SFT 推理管线还使用 Shift 5,以及负分支跳过第 22 层,层号从 0 开始。这两份极简工作流没有单独暴露这两项设置的调节节点。 Base 参数、SFT 参数

    分辨率

    官方推荐约 100 万、150 万或 200 万总像素。公开说明的外推范围约为 25 万至 400 万总像素,宽高均须为 16 的倍数。 分辨率指南

    实用尺寸示例:

    画幅 宽 × 高 约总像素

    方图 1024 × 1024 100 万

    竖图 1024 × 1536 150 万

    宽屏 1936 × 1088 200 万

     

     

    两份工作流当前均设置为 16:9、约 200 万像素。选择本地对应的模型、文本编码器和 VAE,修改 Text (User Prompt) 中的提示词即可生成。使用 Layout 时,将编译后的 JSON 粘贴到同一提示词框,并保持生成尺寸与布局画布一致。

    Description

    Moss image

    Workflows
    Other

    Details

    Downloads
    71
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/23/2026
    Updated
    11/8/2026
    Deleted
    -

    Files

    工作流.zip

    Mirrors

    CivitAI (1 mirrors)