makisekurisu-jp/ComfyUI-Qwen3-VL-GGUF

https://github.com/KLL535/ComfyUI_Simple_Qwen3-VL-gguf

git clone https://github.com/JamePeng/llama-cpp-python
cd llama-cpp-python
git clone https://github.com/ggml-org/llama.cpp ./vendor/llama.cpp
$env:CMAKE_ARGS = "-DGGML_CUDA=on"
D:\ComfyUI\venv\Scripts\python -m pip install -e . --verbose

After compilation is finished, the llama-cpp-python folder must not be deleted.

---

There is no need to compile it manually anymore. You can use JamePeng’s precompiled WHL package, as long as the CUDA version matches exactly.

https://github.com/1038lab/ComfyUI-QwenVL

https://github.com/JamePeng/llama-cpp-python

https://developer.nvidia.com/cuda-toolkit-archive

---

Gemma 4 requires llama‑cpp‑python ≥ 0.3.35

ComfyUI\custom_nodes\ComfyUI_Simple_Qwen3-VL-gguf\system_prompts_user.json

{
    "_system_prompts": {
    },
    "_user_prompt_styles": {
    },
    "_camera_preset": {
    },
    "_model_presets": {
        "gemma-4-E4B-it-IQ4_XS": {
            "model_path": "D:\\ComfyUI\\models\\LLM\\gemma-4-E4B-it-IQ4_XS.gguf",
            "mmproj_path": "D:\\ComfyUI\\models\\LLM\\gemma-4-E4B-it-mmproj-BF16.gguf",
            "output_max_tokens": 2048,
            "ctx": 8192,
            "n_batch": 2048,
            "n_ubatch": 2048,
            "gpu_layers": -1,
            "temperature": 1.0,
            "top_p": 0.95,
            "min_p": 0.01,
            "top_k": 64,
            "repeat_penalty": 1.0,
            "chat_handler": "gemma4",
            "script": "qwen3vl_run.py",
            "silent": false,
            "debug": true,
            "verbose": true,
            "raw_mode": true,
            "prompt_template": "<|turn>system\n{system}<turn|>\n<|turn>user\n{images}\n{user}<turn|>\n<|turn>model\n",
            "stop": ["<turn|>", "<eos>", "<|end_of_turn|>"]
        },
        "Huihui-Qwen3.5-9B-abliterated.i1-IQ4_XS": {
            "model_path": "D:\\ComfyUI\\models\\LLM\\Huihui-Qwen3.5-9B-abliterated.i1-IQ4_XS.gguf",
            "mmproj_path": "D:\\ComfyUI\\models\\LLM\\Qwen3.5-9B-mmproj-BF16.gguf",
            "output_max_tokens": 2048,
            "image_min_tokens": 1024,
            "image_max_tokens": 2048,
            "ctx": 8192,
            "n_batch": 2048,
            "n_ubatch": 512,
            "gpu_layers": -1,
            "temperature": 0.7,
            "top_p": 0.8,
            "min_p": 0.05,
            "top_k": 20,
            "repeat_penalty": 1.0,
            "present_penalty": 1.5,
            "pool_size": 4194304,
            "chat_handler": "qwen35",
            "enable_thinking": false,
            "script": "qwen3vl_run.py",
            "silent": false,
            "debug": true
        },
        "Qwen3-VL-8B-Instruct-IQ4_XS": {
            "model_path": "D:\\ComfyUI\\models\\LLM\\Qwen3-VL-8B-Instruct-IQ4_XS.gguf",
            "mmproj_path": "D:\\ComfyUI\\models\\LLM\\Qwen3-VL-8B-Instruct-mmproj-BF16.gguf",
            "output_max_tokens": 2048,
            "image_min_tokens": 1024,
            "image_max_tokens": 2048,
            "ctx": 8192,
            "n_batch": 2048,
            "n_ubatch": 512,
            "gpu_layers": -1,
            "temperature": 0.7,
            "top_p": 0.92,
            "min_p": 0.01,
            "top_k": 40,
            "repeat_penalty": 1.1,
            "pool_size": 4194304,
            "chat_handler": "qwen3",
            "script": "qwen3vl_run.py",
            "silent": false,
            "debug": true
        }
    }
}