🎨全能视觉风格提示词@炮老师的小课堂

If you've ever wanted a photography director's vocabulary on tap - composition, lens, lighting, color, mood - this node is that, in a dropdown. DapaoLocalVisualStylePromptNode ("全能视觉风格提示词") combines 39 built-in visual style cards, optional real-prompt and source-image retrieval, and a local LLM to output a Chinese+English image prompt. It doesn't generate images - it writes the prompt that makes your image model do what you meant. Another of the pack's seven API-tool ports running on GGUF instead of a key.

The style-card deck is the star. It's organized by visual domain - 产品广告 (暗黑元素奢侈品, 白色画廊高键产品, 动态液体爆破, 产品即世界…), 人像摄影 (安静奢华柔光棚拍, 硬影几何时尚, 皮肤优先美妆近景, 雨夜霓虹电影人像…), plus 绘画与手工, 食物饮品, 插画叙事, 海报封面, 建筑室内, 风景城市, 科幻奇幻, 平面抽象. Pick a card (or auto-select) and the prompt inherits that genre's look.

The controls worth knowing

The required inputs run deep but most default to 自动判断, so the node works if you just type a request:

• 📝 原始视觉需求 - the one main text box. The tooltip carries a real tip: to optimize an old prompt, paste it here and set 创作任务 to 优化现有提示词.
• 🧩 创作任务 - auto / 从零生成文生图提示词 / 优化现有提示词 / 参考图主体保持 / 参考图风格迁移 / 图像编辑与局部修改 / 商品广告主视觉.
• 🔎 真实提示词联网检索 - the interesting one. Turn it on and the node queries a local read-only database and validates source-image URLs online; first use auto-downloads an ~84.6MB database (no image packs). Off by default, and honestly: leave it off unless you specifically want retrieval-augmented prompt writing, because it adds a network dependency to an otherwise fully local node.
• 🎨 视觉风格卡 (39 cards), 🎯 使用场景, 📐 图片比例, 🎯 目标图像模型 (通用/GPT Image 2/Banana/Gemini/Midjourney/FLUX/SD), 🌐 输出语言 (英文推荐/简体中文/中英双份).
• 🎛️ 展开更多视觉控制 - default on; this reveals the advanced panel: 🗂️ 视觉领域, 🧭 构图与视角, 📷 镜头与景深, 💡 光线, 🎨 配色, 🌫️ 画面氛围, 🖼️ 参考图用途, 🔤 画面文字策略, 🔒 核心保留策略, 🚫 避错策略, 🎚️ 风格改造强度, 🧠 细节密度.

The outputs

🎨 最终图像提示词, 🇨🇳 中文提示词, 🇬🇧 英文提示词 (three flavors of the same prompt - take your pick), plus 🧬 Creative Spec视觉规范, 🔎 检索参考报告, 🃏 使用风格卡, and ℹ️ 最终处理信息. The 🖼️ 用户参考图1 through 🖼️ 用户参考图9 optional inputs add your own visual-fact references, which the tooltip notes coexist with (not replace) the online-retrieval references.

Where people get burned

The 🔎 真实提示词联网检索 flag is the trap for the privacy-minded: this pack's whole selling point is "no cloud," and that one toggle reaches the network (for URL validation) plus downloads a database on first use. Fine if you want it, surprising if you didn't notice it. And like all vision nodes here, reference-image analysis needs a matched mmproj on your model.

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/paolaoshi/ComfyUI-llama_Dapao

then requirements + restart. Pack-wide gotchas: llama-cpp-python pinned to the JamePeng fork at 0.3.47+ (stock PyPI 0.3.35 fails on Qwen3.8 GGUFs), NVIDIA users overlay a GPU wheel from the JamePeng releases, models in ComfyUI/models/LLM/.