说清创作意图
Set the intent
描述主体、场景和作品用途，让创作从具体的想法开始。
Describe the subject, scene, and purpose. Begin with a concrete idea.
AIComposer
AI 引导创作与图层图像编辑器
Guided AI creation and layer-based image editing
macOS · ZenMux · Open Source
带上参考图，
把想法变成画面。
Bring your references.
Shape the image you imagine.
从创作意图和风格开始，也可以加入最多 5 张参考图。安排多人造型、换衣与场景，再用 IK 人偶调整姿态；提示词与参考图编号自动同步，生成后继续在原生 Mac 图层编辑器里完成作品。
Start with an idea and a visual style, then add up to five reference images. Direct multiple people, clothing, and scenes, and pose figures with an IK mannequin. Reference numbers stay in sync with the prompt, and the result is ready for layer-based finishing in a native Mac editor.
NEW / Guided creation
先找到表达，再开始生成。
Find the expression. Then generate.
探索视觉风格
Explore a style
按领域或关键词搜索，理解风格特征与模板；也可以自由创作。
Search by category or keyword. Explore traits and templates, or keep your style open.
塑造画面语言
Shape the direction
补充构图、光线、配色和要避免的元素，让模板贴合你的画面。
Add composition, light, color, and elements to avoid. Make the template fit your image.
确认并进入图层
Review and create
编辑组合后的提示词，回到 AI 面板设置模型和尺寸，再继续图层后期。
Edit the assembled prompt, return to the AI panel, then continue finishing with layers.
30 / 13公开风格预览 / 创作领域Public style previews / creative domains
内置风格引用 Vibe Shot Club 风格图谱，每条保留原文链接。来源共 1,617 条，当前内置公开可访问的 30 条；支持导入授权 JSON 风格库。
Style previews are attributed to Vibe Shot Club’s Style Atlas, with original links. The source lists 1,617 entries; 30 publicly accessible previews are bundled. Import an authorized JSON catalog to expand your collection.
使用说明与扩展来源 ↗Guide & expansion sources ↗NEW / References & posing
从图像参考，到人物姿态。
From image references to character poses.
最多 5 张参考图
Up to five references
拖入、粘贴或选择图片，以缩略图管理。每张图可作为灵感、人物脸、服装、姿态或场景参考；删除后提示词自动更新“参考图 1、2、3…”，并按当前顺序发送。
Drop, paste, or choose images and manage them as thumbnails. Assign each as inspiration, identity, clothing, pose, or scene. Removing an image updates the numbered prompt references and sends the remaining images in order.
多人造型与身份保持
Multiple people, distinct identities
为 1–5 位人物分别指定脸部、服装与姿态参考，也可叠加换衣和换场景。换脸时提示模型严格保留参考脸的生物特征，不混合不同人物身份。
Direct one to five people with separate face, clothing, and pose references. Combine outfit and scene changes. Face edits instruct the model to preserve the reference identity’s defining facial features without blending people together.
IK 人偶姿态编辑
IK mannequin posing
用可操控的人偶摆姿势，支持反向运动学（IK）、关节限位、多人独立控制、镜像与自动落地。导出的姿态图作为单独的姿态引导附件，不会和普通参考图编号混淆。
Pose an interactive mannequin with inverse kinematics (IK), joint limits, per-person controls, mirroring, and automatic grounding. The exported pose guide is sent as a separate attachment, distinct from numbered image references.
5,424 / 24公开图库索引 / 归纳画面配方Public gallery index / visual recipes
从 OpenNana 可公开访问的案例样图归纳画面构成、色彩、光线、材质与构图，再融合进创作引导。付费案例不复制原始提示词；根据可见图像提炼视觉特征，让参考灵感服务于你的主题。
Visual traits such as composition, color, lighting, and material are distilled from publicly viewable OpenNana sample images and folded into the creative guide. Paid prompt text is not copied; visible images inform visual direction for your own subject.
AI 强化了什么
What the AI layer adds
生图
Generate
按提示词新建图层：静物、纹理、海报、变体。
Create a new layer from a prompt: still life, texture, poster, or variation.
选区改图
Edit a selection
填充、去物体、去字迹，或把选中的图形文字识别成可编辑文字。原图、选区蒙版和提示词一并送给模型，结果写回当前图层。
Fill, remove objects, clear handwriting, or convert selected graphic type into live editable text. The photo, selection mask, and prompt go out together; the result writes back to the current layer.
扩图与换天空
Expand and sky
生成式扩图、替换天空、风格重绘。会处理接缝、黑边和画幅，避免图被越改越小。
Generative expand, sky replacement, and restyle. Seams, letterboxing, and shrinking frames are handled in the pipeline.
仍然是一台完整的图像编辑器
Still a full image editor
图层、蒙版与 PSD
Layers, masks, and PSD
文件夹、混合模式、不透明度、图层蒙版、剪贴蒙版、调整图层。可导入 Photoshop PSD（8-bit RGB）以及相机 RAW。
Folders, blend modes, opacity, layer masks, clipping masks, and adjustment layers. Import Photoshop PSD (8-bit RGB) and camera RAW.
修饰与变换
Retouch and transform
画笔、修复、仿制、套索、魔棒、扭曲、裁剪，以及色阶/曲线/色相饱和度。
Brush, healing, clone, lasso, magic wand, distort, crop, plus Levels, Curves, and Hue/Saturation.
熟悉的原生界面
A familiar native editor
编辑器支持简体中文、英文、日文、繁体中文、韩文。新版创作引导采用中文界面，风格数据可按系统语言切换。
The editor supports Simplified Chinese, English, Japanese, Traditional Chinese, and Korean. The new creative guide uses a Chinese interface; style metadata follows the system language.
ZenMux 接入
ZenMux inside the app
在设置里填入 ZenMux API Key（加密保存在本机）。生图可选 GPT Image、Qwen Image、Gemini Image；另可填写多模态模型（默认 google/gemini-3.8-flash）用于识别图形文字。没有 Key 时可用邀请链接注册。密钥不会进仓库。
Paste a ZenMux API key in Settings. It is encrypted on this Mac. Image models: GPT Image, Qwen Image, Gemini Image. A separate multimodal model name (default google/gemini-3.8-flash) is used to read graphic type. Invite links work when you do not have a key yet. Keys never enter the repository.
源自 Compositor。 本项目 fork 并修改自 robbietilton/Compositor（MIT）。原作者做了一台免费、开源、面向合成与后期的 Mac 图像编辑器。AIComposer 在此之上加入 ZenMux 生图/改图，并保持原项目的图层工作流。请同时尊重上游许可证与致谢。
Based on Compositor. This project is a fork of robbietilton/Compositor (MIT). The original app is a free, open-source Mac editor for compositing and finishing. AIComposer adds ZenMux generation and editing on that layer workflow. Please keep the upstream license and credit.