Y7 Qwen3-VL

Qwen3-VL<div style="font-size: 0.8em"><div id="Y7_shortdesc" style="margin-bottom: 15px;">Run vision-language inference using Qwen3-VL models. Supports image analysis and text-only queries.</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Uses Qwen3-VL integrated with ComfyUI's memory manager for proper VRAM coordination alongside other loaded models.</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Models:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Qwen/Qwen3-VL-2B-Instruct</code></div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Qwen/Qwen3-VL-4B-Instruct</code></div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Qwen/Qwen3-VL-8B-Instruct</code> (default)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Qwen/Qwen3-VL-32B-Instruct</code></div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Inputs:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">image</code>: Optional image input — omit for text-only queries</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">model_name</code>: Which Qwen3-VL model to use</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">preset_prompt</code>: Built-in instruction for how the model should analyse the image</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">custom_prompt</code>: If filled, replaces the preset instruction entirely</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">max_new_tokens</code>: Maximum tokens to generate (64–4096, default 512)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">temperature</code>: Controls randomness (0.0–2.0, default 0.7)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">top_p</code>: Nucleus sampling threshold (0.0–1.0, default 0.9)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">repetition_penalty</code>: Penalises repeated tokens (1.0–2.0, default 1.1)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">seed</code>: Random seed for reproducibility</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">keep_model_loaded</code>: Keep the model in VRAM between runs to skip reloading</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">download_model</code>: Automatically download the selected model from HuggingFace if not found locally</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Preset Instructions:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Tags</code>: Up to 50 comma-separated visual tags for text-to-image AI</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Simple Description</code>: A single concise sentence about the main subject and setting</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Detailed Description</code>: Paragraph covering subject, environment, lighting, and composition</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Ultra Detailed Description</code>: Extended paragraph with micro-details, textures, and lighting analysis</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Cinematic Description</code>: Film-still style paragraph with camera language and mood</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Detailed Analysis</code>: Structured sections — Subject, People, Environment, Lighting, Camera/Composition, Color/Texture</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Short Story</code>: A short imaginative story inspired by the image</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">Prompt Refine & Expand</code>: Refines and expands a text prompt for creative image generation</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Output:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">response</code>: The model's text response</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Models are stored in <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">models/LLM/<model-name></code>. Enable <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">download_model</code> to download automatically if not found locally.</div></div>