GALIAIS-Nodes 06 场景画面

Characters don't exist in white space, and the 06 场景画面 (Scene & Style) node is where this pack puts everything that isn't the character: camera, location, time of day, props, and the overall visual style. It's the difference between a character study and an image. It also happens to be the layer that benefits most from good prose-plus-tag mix, because scene and style are exactly where an anime-SDXL model's Danbooru training and its world knowledge both show up.

What you fill in

Five content fields plus a supplement, each with a 使用 toggle:

• 镜头构图 (camera/composition) - close-up, full body, from below, rule of thirds, wide shot, the framing tags.
• 地点背景 (location/background) - city street, bedroom, forest, classroom, where this is happening.
• 时间天气 (time/weather) - sunset, night, rain, snowing, golden hour - cheap, high-impact mood.
• 道具生物 (props/creatures) - cell phone, umbrella, butterfly, crowd, the stuff in the frame that isn't the subject.
• 画面风格 (visual style) - watercolor, cel shading, detailed background, glowing - the look of the image itself.
• 场景补充 (scene supplement) - free-text extras, multiline.

Plus the shared controls: 角色槽位 (全局 / 角色1–8), 角色标签, 启用, 权重, and the random/AI block (启用随机Tag, Tag生成模式, 随机种子, optional AI服务商), with optional DB, 上游上下文, and 上游提示词段. Because scene usually applies to the whole image rather than one character, 全局 is the sane default for the slot here.

Outputs

The standard trio: 场景段 (section object → composer), 文本 (plain text), 元信息JSON (contribution + random audit).

Where it earns its keep

The 时间天气 field is the highest-leverage input in the layer - sunset vs rain changes a render more than most other single tags. And 画面风格 is the local, free answer to "why does my output look flat": if you don't want to pay for the pack's AI Style Enhancer, a strong 画面风格 tag here does most of the same work with zero API calls. This layer is also where the random machinery shines most harmlessly - randomizing 时间天气 or 地点背景 across runs is the fastest way to batch variations of the same character in different worlds.

One caution the community will be familiar with: location tags can push lighting (the classic "beach/overexposed" failure). If adding a location makes your render blow out, that's a known association quirk, and the fix is a location-sensitive negative or a weaker location tag - not deleting the whole layer.

The honest read

The pack's AI Scene Director node does fancier things with foreground/midground/background, but for most work this node is the right tool: six fields, local, free, and it feeds straight into the composer. Keep the 画面风格 tags matching your base model's vocabulary (a watercolor tag means nothing to a photographic model), and you're most of the way there.

Install

Pack-wide, once:

git clone https://github.com/GALIAIS/GALIAIS-Nodes.git ComfyUI/custom_nodes/GALIAIS-Nodes

Restart ComfyUI, or ComfyUI Manager → "GALIAIS-Nodes". No pip dependencies; the Danbooru DB is separate if you want dictionary-backed picking. New pack, no community writeups - the 元信息JSON and README flow are your docs.