ComfyUI Extension: ComfyUI-Qwen2.5-VL-7B-OPENVINO
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.
A custom ComfyUI node that unleashes Intel CPU + iGPU power via OpenVINO.
Looking for a different extension?
Custom Nodes (5)
README
⚡ComfyUI-Qwen2.5-VL-7B-OPENVINO
Text-to-Image at Blazing Speed — Powered by Intel OpenVINO
</div>🌟 簡介 / Introduction
為 Intel 內顯玩家量身打造的 ComfyUI 加速節點!
A custom ComfyUI node that unleashes Intel CPU + iGPU power via OpenVINO.
這個節點由 Claude / DeepSeek / Gemini AI 協助開發,讓配備 Intel CPU(如 i5-1135G7)的使用者,能在 ComfyUI 中透過 Intel 內顯(iGPU)執行 Z-Image Turbo OV Model,獲得驚人的生成速度。
| 方案 | 解析度 | Steps | 時間 | |------|--------|-------|------| | 🐢 Z-Image Turbo GGUF Q2 | 512×512 | 7 | ~1500 秒 | | ⚡ 本節點 (OpenVINO) | 512×512 | 7 | 90S 秒 (約快 20 倍!) |
💡 模型作者(非本人): [(https://huggingface.co/helenai/Qwen2.5-VL-7B-Instruct-ov-int4)]
🚀 安裝 / Installation
⚠️ 強烈建議使用全新的虛擬環境 (VENV)
Step 1 — 取得專案 / Clone the repo
cd /ComfyUI/custom_nodes/
git clone https://github.com/blackmeat1225/ComfyUI_Z-Image_turbo_OPENVINO.git
cd /ComfyUI/custom_nodes/ComfyUI_Z-Image_turbo_OPENVINO
model_path=F:\ComfyUI\models\openvino\Qwen2.5-VL-7B-Instruct-ov-int4
Step 2 — 安裝依賴 / Install requirements
⚠️ 必須先執行 uninstall 步驟!
python.exe -m pip install --upgrade pip
pip install openvino-genai pillow
🎨 使用方式 / How to Use
啟動 ComfyUI 後,在左側 Custom Nodes 面板中找到本節點:
<div align="center"> <img src="https://github.com/user-attachments/assets/cb6e847f-9f85-46a5-b243-3a03ce1e409c" width="480" alt="節點面板"/> </div>🟢 Method 1 — Text to Image(純文字生成)
由 Claude AI 設計
直接用文字描述你想要的畫面:
The subject in the image is an anime-style character with long, straight silver hair.
She is wearing a black, form-fitting outfit that includes a lace bra and a matching skirt.
The outfit has a glossy finish, giving it a sleek and elegant look...
<div align="center">
<img src="https://github.com/user-attachments/assets/a90905be-3ec4-47a8-953c-a8aca4191199" width="100%" alt="Method 1 示例"/>
</div>
🟡 Method 2 — Florence2 Combo(偽 ControlNet)
由 Gemini AI 設計
📦 需要額外安裝:
pip install timm
看兩張圖,比較久,但效果穩定,偶有火柴人殘影。
<div align="center"> <img src="https://github.com/user-attachments/assets/b2d22fcd-e312-4d77-b650-0afc3c3a4a5b" width="100%" alt="Method 2 示例"/> </div>🔵 Method 3 — Qwen2.5-VL-7B(偽 Image to Image)
由 DeepSeek AI 設計
一眼看一個 → 把兩張圖先合併成一張再看,速度較快,但偶爾會出雙胞胎 👯
<div align="center"> <img src="https://github.com/user-attachments/assets/cfe5c726-ba71-48ba-9f79-1e8b8d455021" width="100%" alt="Method 3 示例"/> </div>🟣 Method 4 — Qwen2.5-VL-7B(偽 ControlNet)
由 DeepSeek AI 設計
<div align="center"> <img src="https://github.com/user-attachments/assets/66b1850f-21f8-400c-aabf-976c3fe831c0" width="100%" alt="Method 4 示例"/> </div>📋 節點說明 / Node Reference
| 節點名稱 | 功能 | 備註 |
|----------|------|------|
| ZITNT_SIMPLE | 純文字生成圖片 | 🌟 主節點,推薦使用 |
| Florence2 Combo | 圖片引導生成 | 需安裝 timm |
| Qwen2.5-VL 單張 | Image to Image | 偶有雙胞胎問題 |
| Qwen2.5-VL 雙張 | ControlNet 風格 | 速度稍慢但穩定 |
其他節點為 AI 協助生成,
ZITNT_SIMPLE是作者最推薦的核心節點,其他節點可自行探索。
🙏 致謝 / Credits
- 🧠 模型作者: hsuwill000 — Z-Image-Turbo-ov
- 🤖 AI 協作: Claude、DeepSeek、Gemini
- ⚙️ 核心技術: Intel OpenVINO · ComfyUI · HuggingFace Diffusers
<div align="center">
希望讓所有 Intel 內顯玩家都能享受 AI 繪圖的樂趣! 🎉
Made with ❤️ for the Intel iGPU community
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.