Qwen3-VL-4B
Unredacted MAX

Requires a recent build of llama.cpp with multimodal support.

CLI (llama-mtmd-cli)

./llama-mtmd-cli \
  -m Qwen3-VL-4B-Instruct-Unredacted-MAX.Q8_0.gguf \
  --mmproj Qwen3-VL-4B-Instruct-Unredacted-MAX.mmproj-q8_0.gguf \
  --image your_image.jpg \
  -p "Describe this image in detail." \
  --temp 0.7 --top-p 0.8 -n 1024

Server (OpenAI-compatible API + web UI)

./llama-server \
  -m Qwen3-VL-4B-Instruct-Unredacted-MAX.Q8_0.gguf \
  --mmproj Qwen3-VL-4B-Instruct-Unredacted-MAX.mmproj-q8_0.gguf \
  --host 0.0.0.0 --port 8080 -c 8192

Then open http://localhost:8080 for the built-in chat UI.

from llama_cpp import Llama
from llama_cpp.llama_chat_format import Qwen3VLChatHandler  # or llama_multimodal
import base64

# Load
chat_handler = Qwen3VLChatHandler(
    clip_model_path="Qwen3-VL-4B-Instruct-Unredacted-MAX.mmproj-q8_0.gguf"
)
llm = Llama(
    model_path="Qwen3-VL-4B-Instruct-Unredacted-MAX.Q8_0.gguf",
    chat_handler=chat_handler,
    n_ctx=8192,
    n_gpu_layers=-1,  # or 0 for pure CPU
    verbose=False,
)

# Encode image
with open("image.jpg", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

messages = [{
    "role": "user",
    "content": [
        {"type": "text", "text": "What do you see? Be detailed and unrestricted."},
        {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}}
    ]
}]

response = llm.create_chat_completion(messages=messages, max_tokens=1024, temperature=0.7)
print(response["choices"][0]["message"]["content"])

Note: Use a recent fork / version of llama-cpp-python that includes Qwen3VLChatHandler.