想在家用一般中等顯卡做「換裝、改場景」這類高精度圖片編輯？ComfyUI 以節點與流程圖為核心，能把整個推論管線直覺拼裝；而 Qwen-Image-Edit 把輸入圖同時送進 Qwen2.5-VL（語義控制）與 VAE（外觀控制），因此能做到語義與外觀的雙重編輯，還支援中英雙語的精準文字修改（量化版模型效果可能差強人意），非常適合此次示範用之提示詞「把日照改成夕陽、同時換衣服顏色」的需求。

本文在 RTX 3060 12GB 上示範以 GGUF 量化模型＋Lightning LoRA 的四步加速運行 Qwen-Image-Edit。提供完整模型放置路徑、節點與參數建議，以及實測耗時：首次生圖約 2 分多鐘，改提示詞再次生成約 1 分多鐘。若你顯存或 RAM 更充足，也會說明如何改用較大、更高品質的模型，用時間換取畫質。

💬 提醒：匯入本文提供的工作流後，若出現紅色節點或找不到元件，請把缺少的客製化節點安裝齊全並重啟 ComfyUI；本文選用的是量化版本以兼顧速度與記憶體占用，若你的電腦資源足夠，可改抓更大、更高品質的模型。

一、環境與必備元件

安裝 ComfyUI：依各平台指引完成安裝並能啟動 Web 介面。（Windows 使用者看過來！）
安裝自訂節點：
- ComfyUI-GGUF（提供 UnetLoaderGGUF、CLIPLoaderGGUF 等節點）。
- ComfyUI-Essentials（提供 ImageResize+ 等常用節點）。
安裝完務必重啟 ComfyUI，否則節點清單不會更新。
下載並放置模型（本文實測組合）：

模型與路徑

Diffusion（GGUF）：Qwen_Image_Edit-Q3_K_S.gguf ［🔗 下載點］
📁 放置 → ComfyUI/models/diffusion_models/
Text Encoder（GGUF）：原檔 Qwen_Qwen2.5-VL-7B-Instruct-Q3_K_S.gguf，建議改名為
Qwen2.5-VL-7B-Instruct-Q3_K_S.gguf ［🔗 下載點］
📁 放置 → ComfyUI/models/text_encoders/
mmproj（GGUF）：Qwen2.5-VL-7B-Instruct-mmproj-BF16.gguf ［🔗 下載點］
📁 放置 → ComfyUI/models/text_encoders/
VAE：qwen_image_vae.safetensors ［🔗 下載點］
📁 放置 → ComfyUI/models/vae/
LoRA（Lightning 加速）：Qwen-Image-Lightning-4steps-V1.0.safetensors ［🔗 下載點］
📁 放置 → ComfyUI/models/loras/

說明：為了在 12GB VRAM 上維持順暢，本文採用 Q3 量化版。若你顯存或 RAM 充裕，可改用更高位寬或更大檔的模型，並把取樣步數拉高以換取更好的細節。

二、載入工作流與參數建議

載入模型節點：以 UnetLoaderGGUF 載入 Qwen_Image_Edit-Q3_K_S.gguf；CLIPLoaderGGUF 載入 Qwen2.5-VL-7B-Instruct-Q3_K_S.gguf 與 …mmproj-BF16.gguf；VAELoader 載入 qwen_image_vae.safetensors。
匯入圖片：LoadImage 串到 ImageResize+，建議先壓到約 1024×768，再送入編碼／條件節點。
提示詞（關鍵）：用 TextEncodeQwenImageEdit 輸入需求，例如「衣物改白色比基尼，時間改成夕陽時刻」。此節點同時處理語義與外觀約束。
Lightning 套用：以 LoraLoaderModelOnly 載入 Qwen-Image-Lightning-4steps-V1.0，讓 KSampler 可用「4 步」快速生成。
KSampler 設定：steps=4、sampler=euler、scheduler=karras、CFG 建議 1–4；顯存吃緊就維持批量 1、解析度約百萬像素。
輸出：VAEDecode → SaveImage 儲存成果。首次載入模型約 2 分多鐘；同流程改提示詞再跑約 1 分多鐘（RTX 3060 12GB）。

重要：匯入工作流後若有缺節點，請把遺漏的客製化節點安裝完並重啟 ComfyUI，否則無法正常執行。

三、常見問題與調校

顯存不足：先降解析度（例如 1024×768）、保持步數 4、批量 1；必要時改用更小的量化檔。
畫質不足：提高量化等級（如 Q4/Q5）、增加步數或移除 Lightning；但速度與顯存占用會上升。
節點找不到：多半是少裝 ComfyUI-GGUF 或 ComfyUI-Essentials，補齊後重啟。

📝 工作流(ImageEditQwen.min.json)

可以複製回文字編輯器並以 .json 格式儲存以匯入。

{"id":"b1c3aee6-2524-431f-abef-6517a61b2edb","revision":0,"last_node_id":18,"last_link_id":27,"nodes":[{"id":7,"type":"VAEDecode","pos":[-550,-950],"size":[140,46],"flags":{},"order":10,"mode":0,"inputs":[{"name":"samples","type":"LATENT","link":6},{"name":"vae","type":"VAE","link":24}],"outputs":[{"name":"IMAGE","type":"IMAGE","links":[10]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"VAEDecode","widget_ue_connectable":{}},"widgets_values":[]},{"id":8,"type":"ConditioningZeroOut","pos":[-870,-930],"size":[197.712890625,26],"flags":{},"order":8,"mode":0,"inputs":[{"name":"conditioning","type":"CONDITIONING","link":8}],"outputs":[{"name":"CONDITIONING","type":"CONDITIONING","links":[9]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"ConditioningZeroOut","widget_ue_connectable":{}},"widgets_values":[]},{"id":12,"type":"VAEEncode","pos":[-1337.119873046875,-893.5999145507812],"size":[140,46],"flags":{},"order":7,"mode":0,"inputs":[{"name":"pixels","type":"IMAGE","link":20},{"name":"vae","type":"VAE","link":25}],"outputs":[{"name":"LATENT","type":"LATENT","links":[13]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"VAEEncode","widget_ue_connectable":{}},"widgets_values":[]},{"id":6,"type":"KSampler","pos":[-880,-1240],"size":[270,262],"flags":{},"order":9,"mode":0,"inputs":[{"name":"model","type":"MODEL","link":4},{"name":"positive","type":"CONDITIONING","link":5},{"name":"negative","type":"CONDITIONING","link":9},{"name":"latent_image","type":"LATENT","link":13}],"outputs":[{"name":"LATENT","type":"LATENT","links":[6]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"KSampler","widget_ue_connectable":{}},"widgets_values":[707703061584525,"randomize",4,1,"euler","karras",1]},{"id":13,"type":"ImageResize+","pos":[-2030,-1240],"size":[270,218],"flags":{},"order":4,"mode":0,"inputs":[{"name":"image","type":"IMAGE","link":16}],"outputs":[{"name":"IMAGE","type":"IMAGE","links":[19,20]},{"name":"width","type":"INT","links":null},{"name":"height","type":"INT","links":null}],"properties":{"cnr_id":"comfyui_essentials","ver":"9d9f4bedfc9f0321c19faf71855e228c93bd0dc9","Node name for S&R":"ImageResize+","widget_ue_connectable":{}},"widgets_values":[1024,768,"nearest","fill / crop","always",0]},{"id":10,"type":"LoadImage","pos":[-1849.5830078125,-1778.170654296875],"size":[550,450],"flags":{},"order":0,"mode":0,"inputs":[],"outputs":[{"name":"IMAGE","type":"IMAGE","links":[16]},{"name":"MASK","type":"MASK","links":null}],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"LoadImage","widget_ue_connectable":{}},"widgets_values":["test_image.jpg","image"],"color":"#332922","bgcolor":"#593930"},{"id":9,"type":"SaveImage","pos":[-1275.94775390625,-1804.34619140625],"size":[680,480],"flags":{},"order":11,"mode":0,"inputs":[{"name":"images","type":"IMAGE","link":10}],"outputs":[],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"SaveImage","widget_ue_connectable":{}},"widgets_values":["QwenEdit/ComfyUI"],"color":"#232","bgcolor":"#353"},{"id":3,"type":"LoraLoaderModelOnly","pos":[-1342.73193359375,-1236.1573486328125],"size":[300,82],"flags":{},"order":5,"mode":0,"inputs":[{"name":"model","type":"MODEL","link":21}],"outputs":[{"name":"MODEL","type":"MODEL","links":[4]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.49","Node name for S&R":"LoraLoaderModelOnly","widget_ue_connectable":{}},"widgets_values":["Qwen-Image-Lightning-4steps-V1.0.safetensors",1],"color":"#222","bgcolor":"#000"},{"id":14,"type":"UnetLoaderGGUF","pos":[-1734.3046875,-1237.870361328125],"size":[270,58],"flags":{},"order":1,"mode":0,"inputs":[],"outputs":[{"name":"MODEL","type":"MODEL","links":[21]}],"properties":{"cnr_id":"comfyui-gguf","ver":"1.1.4","Node name for S&R":"UnetLoaderGGUF"},"widgets_values":["Qwen_Image_Edit-Q3_K_S.gguf"],"color":"#222","bgcolor":"#000"},{"id":18,"type":"CLIPLoaderGGUF","pos":[-1733.6112060546875,-1128.93505859375],"size":[270,82],"flags":{},"order":2,"mode":0,"inputs":[],"outputs":[{"name":"CLIP","type":"CLIP","links":[27]}],"properties":{"cnr_id":"comfyui-gguf","ver":"1.1.4","Node name for S&R":"CLIPLoaderGGUF"},"widgets_values":["Qwen2.5-VL-7B-Instruct-Q3_K_S.gguf","qwen_image"],"color":"#222","bgcolor":"#000"},{"id":16,"type":"VAELoader","pos":[-1735.0927734375,-1000.87890625],"size":[270,58],"flags":{},"order":3,"mode":0,"inputs":[],"outputs":[{"name":"VAE","type":"VAE","links":[23,24,25]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.57","Node name for S&R":"VAELoader"},"widgets_values":["qwen_image_vae.safetensors"],"color":"#222","bgcolor":"#000"},{"id":5,"type":"TextEncodeQwenImageEdit","pos":[-1350,-1100],"size":[420,150],"flags":{},"order":6,"mode":0,"inputs":[{"name":"clip","type":"CLIP","link":27},{"name":"vae","shape":7,"type":"VAE","link":23},{"name":"image","shape":7,"type":"IMAGE","link":19}],"outputs":[{"name":"CONDITIONING","type":"CONDITIONING","links":[5,8]}],"properties":{"cnr_id":"comfy-core","ver":"0.3.50","Node name for S&R":"TextEncodeQwenImageEdit","widget_ue_connectable":{}},"widgets_values":["衣物改白色比基尼，時間改成夕陽時刻"],"color":"#2a363b","bgcolor":"#3f5159"}],"links":[[4,3,0,6,0,"MODEL"],[5,5,0,6,1,"CONDITIONING"],[6,6,0,7,0,"LATENT"],[8,5,0,8,0,"CONDITIONING"],[9,8,0,6,2,"CONDITIONING"],[10,7,0,9,0,"IMAGE"],[13,12,0,6,3,"LATENT"],[16,10,0,13,0,"IMAGE"],[19,13,0,5,2,"IMAGE"],[20,13,0,12,0,"IMAGE"],[21,14,0,3,0,"MODEL"],[23,16,0,5,1,"VAE"],[24,16,0,7,1,"VAE"],[25,16,0,12,1,"VAE"],[27,18,0,5,0,"CLIP"]],"groups":[],"config":{},"extra":{"ds":{"scale":0.7513148009015778,"offset":[2137.8898449088983,1862.945216733735]},"frontendVersion":"1.25.11","ue_links":[],"links_added_by_ue":[],"VHS_latentpreview":false,"VHS_latentpreviewrate":0,"VHS_MetadataImage":true,"VHS_KeepIntermediate":true},"version":0.4}

▲ 這是 ComfyUI 的工作流截圖：左側載入原圖並以 ImageResize+ 調整成 1024×768；中間以 UnetLoader (GGUF)、CLIPLoader (GGUF)、VAELoader 載入 Qwen-Image-Edit、Qwen2.5-VL 與 VAE，並套用 Lightning 4 步 LoRA；右側 KSampler 設定步數 4、euler + karras，提示詞為「衣物改白色比基尼，時間改成夕陽時刻」，最後用 SaveImage 輸出。

▲ 輸入圖：主角「萌芽娘」在海灘上，白天光感、藍天與海平線都很清楚，畫面中有草帽、防曬乳與墨鏡等小物。此例的目標是保持角色與道具位置不變，僅以文字提示實作「換裝」與「改時間」兩種編輯，以檢驗模型的語義與外觀雙重控制。

▲ 輸出圖：衣著被精準換成白色比基尼，時段也成功轉為夕陽餘暉；沙灘與海面映出橙紅光影，角色膚色與髮絲邊緣過渡自然。個人有用更大的模型跑過，效果會更好！但速度巨慢。另外此工具是可以產生 NSFW 圖像的，請斟酌使用。