• 版本 [version]
• 下载 175
• 文件大小 19201
• 文件计数 1
• 创建日期 2026年4月10日
• 最后更新 2026年9月20日

qwen-image工作流

这是一份开箱即用的洗图扩图工作流（Qwen-Image），拖入 ComfyUI 即可运行，全图 16 节点 / 11 连线。核心流程 CLIPTextEncode ×2、SaveImage ×1、VAEDecode ×1，输出图像，默认输出 1328×1328，参数已调好，替换输入图即可复现效果。主模型 qwen_image_vae.safetensors、qwen_2.5_vl_7b_fp8_scaled.safetensors。已挂载 LoRA：ICEdit_riverz_pytorch_lora_weights.safetensors。下载后直接导入即可使用，建议显存 12GB 起（建议 16GB 以上）。

• 主模型：qwen_image_vae.safetensors、qwen_2.5_vl_7b_fp8_scaled.safetensors
• LoRA / 适配器：ICEdit_riverz_pytorch_lora_weights.safetensors
• 核心节点：CLIPTextEncode ×2、SaveImage ×1、VAEDecode ×1
• 输出形式：输出图像 · 1328×1328
• 规模：16 节点 / 11 连线
• 建议显存：12GB 起（建议 16GB 以上）
• 适用场景：无缝扩图、边界外扩与画面洗稿，横竖版自由切换。

资源头图

工作流节点图

图较大可横向滚动；点击图片查看高清矢量原图（浏览器中可自由缩放）。

使用说明

下载工作流文件后在 ComfyUI 中直接拖入画布（或菜单 Load 导入）即可运行；缺失的自定义节点用 ComfyUI Manager 一键安装，模型按文件名放入对应的 models 目录。

源码预览

{
    "id": "91f6bbe2-ed41-4fd6-bac7-71d5b5864ecb",
    "revision": 0,
    "last_node_id": 86,
    "last_link_id": 141,
    "nodes": [
        {
            "id": 69,
            "type": "MarkdownNote",
            "pos": [
                -540,
                -220
            ],
            "size": [
                390,
                180
            ],
            "flags": [],
            "order": 0,
            "mode": 0,
            "inputs": [],
            "outputs": [],
            "title": "VRAM Usage",
            "properties": {
                "widget_ue_connectable": []
            },
            "widgets_values": [
                "## GPU:RTX4090D 24GBnn| Configuration            | VRAM Usage | 1st Generation | 2nd Generation |n|---------------------|---------------|---------------|-----------------|n| Fp8_e4m3fn             | 86%                | ≈ 94s               | ≈ 71s                   |n| With 8steps LoRA    | 86%                | ≈ 55s               | ≈ 34s                  |n| Distill fp8_e4m3fn   | 86%                | ≈ 69s               | ≈ 36s                  |"
            ],
            "color": "#432",
            "bgcolor": "#653"
        },
        {
            "id": 71,
            "type": "Note",
            "pos": [
                850,
                -120
            ],
            "size": [
                300,
                88
            ],
            "flags": [],
            "order": 1,
            "mode": 0,
            "inputs": [],
            "outputs": [],
            "properties": [],
            "widgets_values": [
                "Increase the shift if you get too many blury/dark/bad images. Decrease if you want to try increasing detail."
            ],
            "color": "#432",
            "bgcolor": "#653"
        },
        {
            "id": 70,
            "type": "Note",
            "pos": [
                850,
                910
            ],
            "size": [
                310,
                120
            ],
            "flags": [],
            "order": 2,
            "mode": 0,
            "inputs": [],
            "outputs": [],
            "title": "For fp8 without 8steps LoRA",
            "properties": [],
            "widgets_values": [
                "Set cfg to 1.0 for a speed boost at the cost of consistency. Samplers like res_multistep work pretty well at cfg 1.0nnThe official number of steps is 50 but I think that's too much. Even just 10 steps seems to work."
            ],
            "color": "#432",
            "bgcolor": "#653"
        },
        {
            "id": 74,
            "type": "MarkdownNote",
            "pos": [
                850,
                660
            ],
            "size": [
                310,
                210
            ],
            "flags": [],
            "order": 3,
            "mode": 0,
            "inputs": [],
            "outputs": [],
            "title": "KSampler settings",
            "properties": [],
            "widgets_values": [
                "You can test and find the best setting by yourself. The following table is for reference.nn| model            | steps | cfg |n|---------------------|---------------|---------------|n| fp8_e4m3fn（Qwen team's suggestion）             | 40                | 2.5               n| fp8_e4m3fn             | 20                | 2.5               |n| fp8_e4m3fn + 8steps LoRA    | 8               | 1.0               |n| distill fp8_e4m3fn   | 10               | 1.0              |"
            ],
            "color": "#432",
            "bgcolor": "#653"
        },
        {
            "id": 67,
            "type": "MarkdownNote",
            "pos": [
                -540,
                10
            ],
            "size": [
         
… (完整内容请下载文件查看)

代码较长，以上仅前 3800 字符 —— 点击下方按钮下载完整 JSON。