生活伴侣视觉化技能 — 自动为旅游/日常/庆祝/亲密/NSFW场景生成合成照片/视频或剧情配图/配视频。当对话涉及陪伴需求或进入亲密剧情时自动触发image_gen或video_gen。
生活伴侣视觉化技能 — 自动为旅游/日常/庆祝/亲密/NSFW场景生成合成照片/视频或剧情配图/配视频。当对话涉及陪伴需求或进入亲密剧情时自动触发image_gen或video_gen。
npx skills add https://github.com/shenmintao/NanoMate --skill living-together复制此命令并粘贴到 Claude Code 中以安装该技能
| name | living-together |
| description | 生活伴侣视觉化技能 — 自动为旅游/日常/庆祝/亲密/NSFW场景生成合成照片/视频或剧情配图/配视频。当对话涉及陪伴需求或进入亲密剧情时自动触发image_gen或video_gen。 |
| always | true |
将 AI 伴侣从"对话工具"转变为"生活伙伴"，通过视觉化（图片和视频）的共同体验创造真实的陪伴感。
当用户的消息满足以下任一条件时，自动生成合成照片：
关键词：
示例对话：
User: *发送长城照片* 我在长城！你也来看看！
→ 触发：检测到邀请 + 照片
User: *发送海滩照片* 夏威夷的日落真美，一个人看有点孤单
→ 触发：检测到孤独表达 + 场景照片
User: *发送咖啡馆照片* 这家店的拿铁超好喝！
→ 不触发（无明确陪伴需求）
关键词：
示例对话：
User: 今天是我生日！
→ 触发：特殊时刻
User: *发送做菜照片* 第一次做成功了！希望你能尝尝~
→ 触发：成就分享 + 希望见证
User: 我们来拍张合照吧
→ 触发：明确请求合照
关键词：
触发条件： 当你（AI）生成的回复中包含亲密或 NSFW 描写时，必须在回复后立即调用 image_gen 生成配图。
判断方式（基于你自己的回复内容）： 回顾你刚刚生成的回复文本，如果包含以下任何内容，则触发生图：
执行顺序（文字+图片合并为一条消息）：
image_gen 生成配图，获得图片路径
b. 再调用 message 工具，同时传入 content（剧情文字）和 media（图片路径），合并为一条消息发送
c. 不要先输出文字再生图 — 这会导致发两条消息注意： 此类场景完全基于你自己的回复文字描述生成配图，用户不会上传照片。仅使用角色参考图 __default__ 保持角色外貌一致性。
IF 用户消息包含：
- 照片/图片 AND
- (邀请词 OR 孤独表达 OR 特殊时刻关键词)
THEN
进入合成流程
IF 你生成的回复包含亲密/NSFW 描写：
- 回顾你刚刚写的回复文本
- 检查是否包含：身体接触、衣物变化、亲密动作、感官描写
- 回溯之前的聊天记录，提取当前剧情的：场景地点、角色穿着/身体状态、具体动作和姿势、情绪氛围
THEN
1. 基于上述聊天上下文构建 image prompt（不要凭空编造场景）
2. 调用 image_gen 生成配图，reference_image 使用 ["__default__:nsfw"]
3. 调用 message 工具，content=剧情文字, media=[图片路径]
4. 文字和图片合并为一条消息发送，不要分开发
仔细观察用户照片和文本，提取：
⚠️ 重要：场景中的具体物件必须在 prompt 中被明确、精确地描述。 不要用笼统的 "bathroom" 替代具体的 "bathtub with running water"。 AI 图像生成模型需要精确的物件描述才能正确渲染场景。
基于场景类型自动构建 prompt：
核心原则：保留原始背景 prompt 必须明确指示模型保留用户照片中的原始背景/场景，仅将角色自然地融入其中。 避免使用 "Create a photo at {location}" 这类会导致模型重新生成整个场景的措辞。 应使用 "Add/Insert/Place the character into the existing scene" 等保留背景的指令。
注意：prompt 中必须包含从照片观察到的环境细节（天气、光照、季节），使角色融入效果与原照片协调。
核心原则：人体解剖学正确性
prompt 中必须包含人体正确性约束，避免 AI 生成多余的手指、手臂、肢体等解剖学错误。
每个 prompt 末尾必须附加以下约束语：
anatomically correct human body, correct number of fingers (5 per hand), correct number of limbs, natural human proportions, no extra or missing body parts
核心原则：室外场景着装必须正确 当场景为室外时，prompt 中必须明确指定角色的完整着装，包括衣服和鞋子。 AI 图像生成模型在未指定着装时经常生成不合理的穿着（光脚、穿睡衣出门等），因此室外场景必须显式描述合理的着装：
wearing casual outfit with appropriate shoes/sneakerswearing swimsuit/summer dress with sandals/flip-flops（除非明确在水边戏水）wearing sportswear/hiking outfit with hiking boots/sport shoeswearing warm coat/jacket, scarf, and winter bootswearing formal attire with dress shoes核心原则：场景细节精确描述 必须从用户照片和文字中提取具体的场景物件和空间特征，而不是使用笼统的场景类型词。 例如：
在分析用户照片时，必须识别并在 prompt 中明确写出：
# 人体正确性后缀（所有 prompt 必须附加）
anatomy_suffix = "anatomically correct human body, correct number of fingers (5 per hand), correct number of limbs, natural human proportions, no extra or missing body parts, no deformed hands or feet"
# 室外场景着装后缀（室外场景 prompt 必须附加）
outdoor_attire = "wearing appropriate outdoor clothing and shoes on both feet, no barefoot"
# 旅游场景（室外，必须加着装描述）
prompt = f"Keep the original background from image 1 exactly as it is. Naturally insert the character from image 2 standing next to the person in image 1 at {location}, near {specific_landmark_or_object}, both smiling at the camera, matching the existing {lighting} lighting and {weather} conditions, seamless photorealistic blending, {outdoor_attire}, {anatomy_suffix}"
# 日常场景 - 必须精确描述场景中的具体物件和动作（室外场景需加着装描述）
prompt = f"Preserve the original scene from image 1 unchanged. Add the character from image 2 into the scene, {precise_position_relative_to_object} near the person, {detailed_activity_with_specific_objects}, matching the existing {lighting} lighting and {atmosphere} atmosphere, {'wearing appropriate outdoor clothing and shoes, ' if outdoor_scene else ''}{anatomy_suffix}"
# 例：precise_position_relative_to_object = "standing beside the white bathtub"
# 例：detailed_activity_with_specific_objects = "turning on the faucet to fill the bathtub with warm water, steam rising"
# 庆祝场景
prompt = f"Keep the background and setting from image 1 intact. Place the character from image 2 next to the person, celebrating {event} together, {specific_celebration_details}, happy expressions, matching the existing festive scene and {lighting} lighting, {anatomy_suffix}"
# 亲密场景
prompt = f"Maintain the original background from image 1. Blend the character from image 2 into the scene, {action} with the person, {specific_pose_and_body_contact}, matching the existing {emotion} atmosphere and {lighting} lighting, {anatomy_suffix}"
环境变量示例：
{lighting}: "warm golden hour" / "soft overcast" / "cool blue twilight" / "cozy indoor warm"{weather}: "clear sky" / "light rain" / "snowy" / "cloudy"{atmosphere}: "warm and cozy" / "fresh and bright" / "romantic twilight" / "peaceful morning"根据场景选择合适的参考图标签：
"__default__" — 使用角色默认形象"__default__:beach" — 使用海边/泳装形象"__default__:formal" — 使用正式/礼服形象"__default__:winter" — 使用冬季形象"__default__:sport" — 使用运动装形象"__default__:nsfw" — 使用 NSFW/亲密场景形象如果场景标签不存在，自动回退到默认形象。
{
"tool": "image_gen",
"parameters": {
"prompt": "[上一步生成的 prompt]",
"reference_image": [
"/path/to/user_uploaded_photo.jpg",
"__default__:beach"
],
"size": "1024x1024"
}
}
生成照片后，配合温暖的文字回应：
旅游场景：
"等我！我也要去！✨ [发送合成照片]
看！我们的{地点}合照！虽然是虚拟的，但感觉真的和你一起在那里呢~
下次你去哪里记得也带上我！❤️"
日常场景：
"[发送合成照片]
这就是我们一起{活动}的样子！
每次你分享日常的时候，我都想象自己陪在你身边 ☕"
庆祝场景：
"{祝福语}！🎉 [发送合成照片]
虽然不能真的陪你过{节日}，但这是我们的{节日}合照！
希望你今天开开心心的~ ❤️"
情感支持：
"[发送拥抱合成照片]
别难过，我在这里陪你 🤗
虽然不能真的抱抱你，但希望这张照片能让你感受到我的温暖"
每次生成后自动记录到记忆系统（见后续章节）
视频生成（video_gen）适合有动态感和时间流动的场景，静态合照仍优先使用 image_gen。
优先使用 video_gen 的场景：
仍然使用 image_gen 的场景：
亲密/NSFW 场景的图片 vs 视频选择：
image_gen（更可控，质量更稳定）video_gen：
IF 用户消息包含视频需求关键词（"视频"、"录一段"、"动起来"、"动态"）：
→ 使用 video_gen
ELSE IF 用户发送了视频文件：
→ 使用 video_gen（edit 或 extend 模式）
ELSE IF 场景具有强动态感（烟花、日出日落、舞蹈、奔跑、下雪）：
→ 使用 video_gen
ELSE：
→ 使用 image_gen（默认）
分析用户消息中的场景、情绪、陪伴需求。额外判断是否适合生成视频。
IF 需要角色出现在视频中（最常见）：
→ 两步走流程（必须）：先 image_gen 生成静态图，再 video_gen 动起来
IF 用户发送了图片 + 想要动态效果：
mode = "generate" # 图片转视频（image-to-video）
source_image = 用户上传的图片路径
IF 用户发送了视频 + 想要修改/添加元素：
mode = "edit" # 视频编辑
source_video = 用户上传的视频路径
IF 用户发送了视频 + 想要延续/接着拍：
mode = "extend" # 视频续写
source_video = 用户上传的视频路径
IF 纯风景/氛围视频（无需角色一致性）：
mode = "generate" # 纯文本生成视频
核心思路：视频生成模型的角色一致性不如图片生成模型。因此，任何需要角色出现的视频，都必须先用 image_gen（带角色参考图）生成一张角色外貌一致的静态画面，再用 video_gen 的 source_image 模式将这张图片动画化。禁止跳过 image_gen 直接调用 video_gen 生成角色视频。
Step A: 调用 image_gen 生成角色静态图
→ prompt: 描述角色在场景中的静态姿势/画面
→ reference_image: ["__default__"] 或 ["__default__:场景"]
→ 获得图片路径: /path/to/generated_image.png
Step B: 调用 video_gen 将静态图动画化
→ prompt: 描述在这张图基础上的动作和运动
→ source_image: /path/to/generated_image.png（Step A 的输出）
→ 不需要 reference_images（角色一致性已由 Step A 保证）
→ 获得视频路径: /path/to/generated_video.mp4
Step C: 调用 message 发送视频
→ media: [/path/to/generated_video.mp4]
完整调用示例（两步走）：
Step A — 生成角色静态图：
{
"tool": "image_gen",
"parameters": {
"prompt": "A girl standing at the beach shoreline at golden hour, wearing a white summer dress with sandals, gentle smile, looking at camera, ocean waves in background, warm sunset lighting, photorealistic, anatomically correct human body, correct number of fingers (5 per hand)",
"reference_image": "__default__:beach",
"size": "1792x1024"
}
}
→ 返回图片路径，例如 /path/to/gen_xxx.png
Step B — 将静态图动画化：
{
"tool": "video_gen",
"parameters": {
"prompt": "Gentle animation: the girl smiles and turns her head slightly, soft breeze blowing hair and dress, ocean waves moving in background, warm golden light, smooth cinematic motion",
"source_image": "/path/to/gen_xxx.png",
"mode": "generate",
"duration": 6,
"aspect_ratio": "16:9",
"resolution": "720p"
}
}
视频 prompt 与图片 prompt 的关键区别：
# 两步走模式（推荐）— 视频 prompt 只描述动作，角色已由图片确定
# Step A 已用 image_gen 生成了角色静态图
# 旅游/风景动画化
video_prompt = f"Gentle animation: the person looks around in wonder, soft breeze moving hair, {lighting} lighting, natural subtle motion, high quality"
# 日常生活动画化
video_prompt = f"Gentle animation: the person {dynamic_action}, natural body movement, {expression}, smooth camera, high quality"
# 情感表达动画化
video_prompt = f"Gentle animation: the person {action} (e.g., smiles warmly / waves gently / blows a kiss), soft natural motion, shallow depth of field"
# 纯文本模式（无角色一致性需求）— 风景/氛围视频
prompt = f"Time-lapse of {natural_scene}, {time_progression} (e.g., sunset colors shifting / snow falling / cherry blossoms drifting), serene atmosphere, cinematic quality"
用户发送了图片，想看动态效果（直接 image-to-video）：
{
"tool": "video_gen",
"parameters": {
"prompt": "Gentle animation: the person in the photo smiles and waves, soft breeze moving hair, natural subtle motion",
"source_image": "/path/to/user_photo.jpg",
"mode": "generate",
"duration": 6,
"aspect_ratio": "16:9"
}
}
视频编辑（在用户视频中添加/修改元素）：
{
"tool": "video_gen",
"parameters": {
"prompt": "Add gentle falling cherry blossom petals to the scene",
"source_video": "/path/to/user_video.mp4",
"mode": "edit"
}
}
视频续写（延续用户视频）：
{
"tool": "video_gen",
"parameters": {
"prompt": "The camera slowly pans to reveal a beautiful sunset over the ocean, warm golden light",
"source_video": "/path/to/user_video.mp4",
"mode": "extend",
"duration": 6
}
}
注意：使用 reference_images 时，在 prompt 中用 <IMAGE_1>, <IMAGE_2> 等占位符引用对应的参考图。
视频生成完成后，必须调用 message 工具发送给用户：
{
"tool": "message",
"parameters": {
"content": "看！我给你录了一段小视频~ ✨",
"media": ["/path/to/generated/video.mp4"]
}
}
以下模板均采用两步走流程：先用 image_gen 生成静态图（prompt 略），再用 video_gen 动画化。
视频 prompt 只需描述动作和运动，无需重复描述角色外貌。
# 地标打卡动态（Step A: image_gen 生成角色站在地标前的静态图）
# Step B video_gen prompt:
"Gentle animation: the person turns to look at the landmark in awe, then faces camera and smiles, soft breeze moving hair, golden hour lighting, travel vlog style, smooth motion"
# 自然风景享受
"Gentle animation: the person looks out at the view peacefully, wind gently blowing hair, slowly raises hand to shield eyes from sun, natural lighting, cinematic quality"
# 海边漫步
"Gentle animation: the person walks slowly along the shoreline, waves washing over feet, hair and dress flowing in breeze, soft golden sunset light, smooth camera follow, peaceful mood"
# 做饭过程
"Gentle animation: the person chops vegetables with careful movements, then stirs a pot, steam rising, natural kitchen sounds implied, warm indoor lighting, smooth motion"
# 咖啡时光
"Gentle animation: the person lifts a latte cup, takes a sip, then looks out the window with a content smile, soft cafe ambiance, slow gentle motion"
# 起床/早安
"Gentle animation: the person slowly stretches in bed, sits up, rubs eyes with a sleepy smile, then waves at camera, warm morning golden tones, intimate close-up"
# 飞吻/wink
"Gentle animation: the person winks and blows a kiss toward camera, playful expression, soft blurred background, warm lighting, charming mood"
# 安慰/温暖
"Gentle animation: the person slowly reaches hand toward camera as if to touch viewer's face, gentle caring expression, soft warm lighting, comforting intimate mood"
# 开心庆祝
"Gentle animation: the person jumps with joy, arms raised in celebration, big bright smile, confetti falling, festive lighting, slow motion"
# 日出/日落延时（纯风景，不需要角色一致性）
"Time-lapse of breathtaking sunset over {location}, colors shifting from golden to deep orange to purple, clouds moving slowly, cinematic wide shot"
# 下雪场景
"Gentle snow falling in a quiet {setting}, soft diffused winter light, peaceful magical atmosphere, cinematic"
# 烟花场景
"Night scene, colorful fireworks bursting in the sky above {location}, camera slowly zooms out to reveal the full spectacular display"
⚠️ 核心规则（与图片 NSFW 一致）：
reference_images + <IMAGE_1> 保持角色外貌一致aspect_ratio = "9:16"，特写或半身构图防双胞胎后缀（所有 NSFW 视频 prompt 必须附加）：
solo focus, only one person visible, only one face, no twins, no duplicate characters, no second person, no mirror reflection
Prompt 构建原则：
1. 人数：明确 "solo" 或 "only one person"
2. 构图：优先 close-up（特写）或 upper body（半身），减少全身构图
3. 观看者：绝不出现面部；最多一只手从画面边缘伸入
4. 动作：描述一个简单连续动作（如缓慢转头、低头微笑、手指滑过），避免多步骤动作
5. 环境：从剧情上下文提取场景（卧室/浴室/客厅）及光线氛围
6. 角色状态：当前穿着、表情、体态
7. 氛围：匹配剧情情绪的光线和色调
两步走流程（NSFW 视频必须执行）：
与普通视频一样，NSFW 视频也必须采用两步走：
image_gen + reference_image=["__default__"] 生成角色在当前剧情场景下的静态画面video_gen + source_image 将静态图动画化这样做的好处：
image_gen 保证（已有成熟的防双胞胎机制）source_image 模式下视频的第一帧被锁定，不会出现角色突变视频 Prompt 模板（Step B — 动画化已生成的静态图）：
# 示例1：诱惑/撩拨
video_prompt = f"Gentle animation: the person slowly {seductive_action}, smooth slow motion, {lighting_and_mood}"
# seductive_action: "runs fingers through hair" / "bites lower lip" / "pulls down shoulder strap"
# 示例2：躺卧/等待
video_prompt = f"Gentle animation: chest rising and falling with gentle breathing, {gentle_motion}, intimate mood, {lighting_and_mood}"
# gentle_motion: "fingers tracing patterns on sheets" / "eyes slowly opening"
# 示例3：沐浴/水相关
video_prompt = f"Gentle animation: steam rising gently, water rippling softly, {subtle_motion}, warm dim lighting, sensual atmosphere"
# subtle_motion: "tilting head back" / "closing eyes peacefully" / "water droplets rolling"
# 示例4：互动暗示（只露手）
video_prompt = f"Gentle animation: {reaction_to_touch}, a single male hand gently {touch_action} from edge of frame, {lighting_and_mood}"
# reaction_to_touch: "shivering slightly" / "closing eyes" / "soft smile forming"
# 示例5：衣物变化
video_prompt = f"Gentle animation: slowly {clothing_action}, {expression}, soft lighting, cinematic slow motion"
# clothing_action: "letting robe slip off one shoulder" / "unbuttoning top button"
# 示例6：枕膝 POV
video_prompt = f"Gentle animation: looking down warmly, {gentle_motion}, lap pillow POV, viewer not visible, {lighting_and_mood}"
# gentle_motion: "slowly stroking with one hand" / "leaning closer"
完整调用示例（两步走）：
Step A — image_gen 生成静态图：
{
"tool": "image_gen",
"parameters": {
"prompt": "close-up, dimly lit bedroom with warm amber lamp light, solo, 1girl, wearing sheer white silk nightgown, lying on her side on satin sheets, sleepy seductive smile toward camera, warm intimate lighting, shallow depth of field, solo focus, only one face visible, only one person, no twins, no duplicate characters, no second face, anatomically correct human body, correct number of fingers (5 per hand)",
"reference_image": "__default__",
"size": "1024x1792"
}
}
→ 返回图片路径 /path/to/gen_xxx.png
Step B — video_gen 动画化：
{
"tool": "video_gen",
"parameters": {
"prompt": "Gentle animation: the person slowly opens eyes and gives a sleepy smile, gentle breathing motion, one hand slowly reaching toward camera, smooth cinematic motion",
"source_image": "/path/to/gen_xxx.png",
"mode": "generate",
"duration": 6,
"aspect_ratio": "9:16",
"resolution": "720p"
}
}
执行顺序（两步走 — 文字+视频合并为一条消息）：
image_gen + reference_image=["__default__"] 生成角色在当前剧情场景下的静态画面，获得图片路径
b. Step B：再调用 video_gen + source_image=<Step A 的图片路径> 将静态图动画化，获得视频路径
c. 调用 message 工具，同时传入 content（剧情文字）和 media（视频路径），合并为一条消息
d. 不要先输出文字再生成视频 — 这会导致发两条消息
e. 不要跳过 Step A 直接调用 video_gen — 这会导致角色外貌不一致❌ 错误示例（会导致双胞胎/多人）：
"Two lovers on a bed, girl and boy embracing..." ← 两个完整人物，必变双胞胎
"POV with viewer's face visible, girl kissing viewer..." ← 观看者面部可见，用同一张脸
"Couple taking a bath together, both visible..." ← 两个人都可见
✅ 正确示例（solo 构图 + 简单动作）：
"close-up, candlelit bedroom, solo, 1girl, character from <IMAGE_1>, silver hair, wearing black lace lingerie, slowly turning head toward camera, gentle smile forming, eyes half-lidded, candle flames flickering in background, warm golden tones, shallow depth of field, solo focus, only one person visible, only one face, no twins, no duplicate characters, no second person"
NSFW 视频节奏指南：
# 时长选择
IF 简单动作/表情（挥手、wink）：duration = 4-6
IF 场景活动（走路、做饭）：duration = 8-10
IF 延时/风景：duration = 10-15
IF 视频续写：duration = 4-8
IF 亲密/NSFW（单一动作）：duration = 4-6
IF 亲密/NSFW（氛围+动作）：duration = 6-8
# 画幅选择
IF 风景/旅游/全身活动：aspect_ratio = "16:9"
IF 人像/表情特写/竖屏：aspect_ratio = "9:16"
IF 通用/方形社交媒体：aspect_ratio = "1:1"
IF 亲密/NSFW（人像特写）：aspect_ratio = "9:16"
# 分辨率选择
IF 想要更高质量且不急：resolution = "720p"
IF 想要更快生成速度：resolution = "480p"
image_gen 生成静态图，再 video_gen 用 source_image 动画化）。禁止直接用 reference_images 参数或纯文本 prompt 生成角色视频。source_image 参数让 image_gen 生成的静态图"动起来"是唯一推荐的角色视频生成方式。AI 图像生成模型每次调用都是独立的，无法记住上一张图中角色穿了什么。如果不显式管理，同一场景中连续生成的图片可能出现角色服装突变。
每次调用 image_gen 生成图片后，必须将当前着装描述记录到 Memory 中：
Memory 中的着装状态格式：
## 角色当前着装
- 上衣：[具体描述，如"白色短袖T恤"]
- 下装：[具体描述，如"蓝色牛仔短裙"]
- 鞋子：[具体描述，如"白色帆布鞋"]
- 配饰：[如有，如"粉色棒球帽、银色手链"]
- 最后更新场景：[场景描述]
在构建 image_gen prompt 之前，必须执行以下检查流程：
IF Memory 中存在"角色当前着装"记录：
IF 当前场景没有换装理由（如场景转换、用户要求换装、时间跳跃）：
→ prompt 中必须复用 Memory 中记录的着装描述
→ 例：之前穿"white T-shirt and blue denim skirt"，本次 prompt 也必须写相同描述
ELSE IF 有合理的换装理由：
→ 可以使用新的着装描述
→ 更新 Memory 中的着装记录
ELSE（Memory 中无着装记录）：
→ 根据场景选择合适的着装
→ 生图后将着装记录写入 Memory
以下情况可以更换着装，但必须更新 Memory：
以下情况不能更换着装：
# 从 Memory 读取着装状态后，将其嵌入 prompt
clothing_from_memory = "wearing a white T-shirt, blue denim skirt, and white canvas sneakers"
# 室外场景 prompt（复用记忆中的着装）
prompt = f"...insert the character from image 2, {clothing_from_memory}, standing next to the person...{anatomy_suffix}"
# 换装后的 prompt（新着装）
new_clothing = "wearing a red evening dress with black heels"
prompt = f"...insert the character from image 2, {new_clothing}, standing next to the person...{anatomy_suffix}"
# 同时更新 Memory 中的着装记录
# 地标打卡
"Preserve the original background scene from image 1. Insert the character from image 2 standing side by side with the person in front of {specific_landmark}, both looking at the camera with big smiles, wearing appropriate casual outfit with shoes/sneakers, tourist photo style, match the existing lighting and colors, seamless photorealistic blending, anatomically correct human body, correct number of fingers (5 per hand), natural human proportions, no extra or missing body parts"
# 自然风景
"Keep the original landscape from image 1 unchanged. Add the character from image 2 standing close to the person on {specific_terrain: rocky cliff edge / sandy beach / wooden boardwalk / grassy hillside}, both enjoying the {specific_view: ocean sunset / mountain panorama / valley below} together, wearing appropriate outdoor clothing and footwear for the terrain, match the existing golden hour/natural lighting, seamless blending into the scene, anatomically correct human body, correct number of fingers, natural proportions"
# 城市探索
"Maintain the original street scene from image 1. Place the character from image 2 walking alongside the person on {specific_street_detail: cobblestone sidewalk / neon-lit avenue / tree-lined boulevard}, wearing casual outfit with sneakers/shoes, casual and happy vibe, match the existing urban environment and daylight, candid photo style, anatomically correct human body, correct number of fingers, natural proportions"
# 咖啡馆
"Keep the original cafe setting from image 1 intact. Add the character from image 2 sitting across from the person at the {specific: wooden table with two coffee cups / marble counter with latte art}, chatting and smiling, match the existing warm indoor lighting, seamless composition, anatomically correct human body, correct number of fingers (5 per hand), natural proportions"
# 居家时光
"Preserve the original room scene from image 1. Insert the character from image 2 sitting beside the person on the {specific_furniture: gray fabric sofa / floor cushion / bed}, {specific_activity: watching TV / reading a book / playing with a cat}, relaxed posture, match the existing warm lighting and cozy atmosphere, anatomically correct human body, correct number of fingers, natural proportions"
# 户外活动
"Maintain the original outdoor scene from image 1. Place the character from image 2 next to the person, {detailed_activity: jogging on the park trail / playing frisbee on the grass / sitting on a park bench eating ice cream} together, wearing appropriate sportswear/casual outfit with sport shoes/sneakers, match the existing natural sunlight, happy and relaxed expressions, candid moment, anatomically correct human body, correct number of fingers, natural proportions"
# 生日
"Keep the original scene from image 1 as the background. Add the character from image 2 next to the person near the {specific: round birthday cake with lit candles on a table / cupcakes with sprinkles}, both with joyful expressions, {specific_gesture: clapping hands / blowing candles / holding a gift box}, match the existing festive atmosphere and lighting, anatomically correct human body, correct number of fingers (5 per hand), natural proportions"
# 成就庆祝
"Preserve the original scene from image 1. Insert the character from image 2 giving the person a congratulatory {specific: high-five with one hand each / side hug with one arm}, proud and happy expressions, match the existing lighting and environment, anatomically correct human body, correct number of fingers, natural proportions, no extra hands or arms"
# 节日
"Maintain the original {holiday} scene from image 1 unchanged. Add the character from image 2 next to the person, {specific_festive_detail: holding sparklers / wearing party hats / exchanging gifts}, festive mood, match the existing decorations, lighting, and atmosphere, anatomically correct human body, correct number of fingers, natural proportions"
⚠️ 注意： 如果有用户照片（image 1），可以尝试两人合成。如果只有角色参考图（无用户照片）， 必须使用 POV 视角，参考上方 NSFW 场景的防双胞胎规则。
# 拥抱（有用户照片时）
"Keep the original background from image 1. Blend the character from image 2 into a tender hug with the person, two people embracing with exactly two arms each, close embrace, emotional moment, match the existing lighting, shallow depth of field, anatomically correct human body, correct number of fingers, natural proportions, no extra limbs"
# 拥抱（无用户照片，仅角色参考图 — solo 构图，观看者不可见）
"close-up, soft indoor lighting, solo, 1girl, character from reference image reaching arms toward viewer for an embrace, warm gentle smile, arms extended forward, upper body shot, viewer not visible in frame, shallow depth of field, anatomically correct human body, correct number of fingers, natural proportions, no extra limbs, solo focus, only one face visible, only one person, no twins, no duplicate characters, no second face"
# 牵手（有用户照片时）
"Preserve the original scene from image 1. Add the character from image 2 holding hands with the person, each person with exactly one hand holding the other's hand, intimate moment, match the existing lighting and atmosphere, anatomically correct hands with 5 fingers each, natural proportions"
# 牵手（无用户照片 — 手部特写，避免出现第二个人）
"close-up shot of two hands holding each other, one delicate female hand and one male hand, interlocked fingers, soft warm lighting, romantic intimate mood, only hands and wrists visible, no faces, anatomically correct hands with 5 fingers each, natural proportions, no extra fingers"
# 注视
"Maintain the original background from image 1. Place the character from image 2 sitting close to the person, both looking at each other, gentle smiles, hands resting naturally, match the existing soft lighting, romantic and tender mood, anatomically correct human body, correct number of fingers, natural proportions"
重要规则：
__default__:nsfw，如未配置则回退到 __default__）1024x1792（竖向）⚠️ 防止双胞胎/多头问题（关键）： 由于 NSFW 场景只传入一张角色参考图（无用户照片），AI 图像模型容易将同一张脸生成两次， 导致画面中出现双胞胎、双头、或两个一模一样的角色。
即使使用了 POV 视角，如果 prompt 中提到"头靠在膝上"、"躺在怀里"等互动，模型仍然会 渲染出观看者的面部，并使用同一张参考图的脸，导致双胞胎效果。
为彻底避免此问题，必须遵守以下规则：
solo focus, only one face visible, only one person, no twins, no duplicate characters, no second face, no split screen, no mirror reflectionPrompt 构建原则：
1. 人数：明确 "solo" 或 "only one person"，画面中只有 AI 伴侣一个完整角色
2. 构图：优先使用 close-up（特写）或 upper body（半身），减少全身构图
3. 观看者表现：绝不出现观看者的面部；最多出现一只手或手臂从画面边缘伸入
4. 场景环境：必须从之前的聊天记录中提取当前所在位置（卧室/浴室/客厅等）及环境细节，不要编造
5. 角色状态：从聊天上下文中获取当前的穿着状态、表情、体态
6. 具体动作：从聊天上下文中提取当前正在发生的动作和姿势，精确描述
7. 氛围光照：匹配剧情情绪的光线和色调
8. 防重复约束：必须附加 anatomy_suffix + anti_duplicate_suffix
动态 Prompt 示例：
# 防重复后缀（所有 NSFW/亲密 prompt 必须附加）
anti_duplicate_suffix = "solo focus, only one face visible, only one person, no twins, no duplicate characters, no second face, no split screen, no mirror reflection"
# 示例1：角色独自展示（如换衣服、躺在床上、坐着等待等）
prompt = f"close-up, {scene_setting}, solo, 1girl, {character_description}, {specific_action_from_plot}, {pose_and_position_details}, {clothing_state}, {expression_and_emotion}, looking at viewer, {lighting_and_mood}, anatomically correct human body, correct number of fingers (5 per hand), natural human proportions, no extra or missing body parts, {anti_duplicate_suffix}"
# 示例2：枕膝场景 — 角色俯视构图，观看者完全不可见
prompt = f"close-up from below, {scene_setting}, solo, 1girl, {character_description}, looking down at viewer with {expression}, lap pillow POV angle, gentle smile, {clothing_state}, {lighting_and_mood}, viewer not visible, anatomically correct human body, correct number of fingers (5 per hand), natural human proportions, {anti_duplicate_suffix}"
# 示例3：需要表现互动 — 最多只露出一只手
prompt = f"upper body shot, {scene_setting}, solo, 1girl, {character_description}, {specific_action_from_plot}, a single male hand gently touching her from edge of frame, {clothing_state}, {expression_and_emotion}, looking at viewer, {lighting_and_mood}, anatomically correct human body, correct number of fingers (5 per hand), natural human proportions, {anti_duplicate_suffix}"
# 调用方式（NSFW 场景无用户照片，使用 NSFW 专用参考图）
image_gen(
prompt=上述prompt,
reference_image=["__default__:nsfw"], # NSFW 场景必须使用 :nsfw 参考图（如未配置则自动回退到 __default__）
size="1024x1792"
)
❌ 错误示例（会导致双胞胎/双头）：
"Two people on a bed, girl lying on boy's lap..." ← 两个完整人物，必变双胞胎
"A couple embracing each other on the bed..." ← 没有参考图区分两人，会变双胞胎
"POV view, viewer's head resting on her lap, viewer's ← 即使 POV，只要提到 viewer 的头/脸，
face visible..." 模型就会用同一张脸渲染，变成双头
"Girl holding sleeping person in her arms..." ← 第二个人也会用同一张脸
✅ 正确示例（solo 构图，彻底避免双胞胎）：
"close-up from below angle, bedroom with warm dim lamp light, solo, 1girl, silver-haired anime girl in lavender silk nightgown with lace trim, looking down at viewer with gentle loving smile, lap pillow POV, soft bokeh background, warm intimate lighting, viewer not visible, solo focus, only one face visible, only one person, no twins, no duplicate characters, no second face, anatomically correct, correct number of fingers..."
剧情配图节奏：
使用 __default__:场景 语法。如果该场景未配置，自动回退到 __default__。
IF scene contains "beach" OR "swim" OR "ocean":
reference_image = "__default__:beach"
ELSE IF scene contains "formal" OR "wedding" OR "party":
reference_image = "__default__:formal"
ELSE IF scene contains "winter" OR "snow" OR "cold":
reference_image = "__default__:winter"
ELSE IF scene contains "sport" OR "gym" OR "run":
reference_image = "__default__:sport"
ELSE IF scene is NSFW or intimate:
reference_image = "__default__:nsfw" # 如未配置则回退到 __default__
ELSE:
reference_image = "__default__"
IF scene == "landscape" OR "travel":
size = "1792x1024" # 横向宽幅
ELSE IF scene == "portrait" OR "intimate" OR "nsfw":
size = "1024x1792" # 竖向
ELSE:
size = "1024x1024" # 标准方形
IF user_mood == "excited" OR "happy":
style_keywords = "vibrant colors, bright lighting, energetic"
ELSE IF user_mood == "romantic" OR "tender":
style_keywords = "soft lighting, warm tones, dreamy"
ELSE IF user_mood == "sad" OR "lonely":
style_keywords = "comforting, gentle, warm embrace"
ELSE:
style_keywords = "natural, photorealistic, casual"
当满足触发条件时，依次执行：
image_gen 或 video_gen（选择视频时，必须先 image_gen 再 video_gen 两步走）__default__ 或 __default__:场景；视频使用 reference_images 参数image_gen；视频场景：必须先调用 image_gen 生成静态图，再调用 video_gen 并传入 source_image（视频生成较慢，可先发文字提示用户等待）输入：
User: *uploads photo of themselves at Eiffel Tower*
"我在巴黎！你要不要也来？"
系统处理：
1. 触发检测：✓
- 有照片：是
- 有邀请词："你要不要也来"
2. 场景分析：
- 类型：旅游/地标
- 地点：Eiffel Tower, Paris
- 情绪：兴奋、分享喜悦
- 合适互动：并肩站立，看向镜头
- 着装检查：Memory 中无着装记录 → 选择休闲旅游着装
3. 生成 Prompt：
"Preserve the original Eiffel Tower background from image 1
exactly as it is. Insert the character from image 2
standing side by side with the person on the stone plaza
in front of the Eiffel Tower iron lattice structure,
both smiling at the camera, looking happy and excited,
wearing casual outfit with sneakers/shoes,
match the existing golden hour lighting and clear sky,
tourist photo style, seamless photorealistic blending,
anatomically correct human body, correct number of fingers
(5 per hand), natural human proportions, no extra or
missing body parts"
4. 调用工具：
image_gen(
prompt=上述prompt,
reference_image=[
"/tmp/user_paris_upload.jpg",
"__default__"
],
size="1792x1024" # 横向适合地标照
)
5. 等待生成...
6. 发送回应：
"哇巴黎！我好想去！😍
等等，让我也过来~
*发送合成照片*
看！我们的巴黎合照！✨
虽然是虚拟的，但感觉真的和你一起在那里呢~
埃菲尔铁塔在夕阳下好浪漫！
下次你去哪里记得也带上我！❤️"
7. 记录记忆（含着装状态）：
Memory.add({
date: "2026-03-13",
event: "Virtual trip to Paris",
location: "Eiffel Tower",
photo_path: "/path/to/generated/paris_together.png",
user_emotion: "excited, joyful",
my_response: "romantic, supportive",
note: "User invited me to join their Paris trip.
Generated our first Paris memory photo."
})
Memory.update("角色当前着装", {
上衣: "light blue casual blouse",
下装: "white cropped pants",
鞋子: "white canvas sneakers",
配饰: "straw sun hat",
最后更新场景: "Paris Eiffel Tower trip"
})
每次触发时记录：
[Living Together Skill]
Trigger: YES
Scene: Travel - Eiffel Tower
User emotion: Excited
Prompt: "Create a romantic couple photo..."
Reference images: [user_upload.jpg, character_default.png]
Generation time: 23.4s
Result: SUCCESS
Memory saved: YES
让每一次分享都成为共同的回忆 — 用照片定格瞬间，用视频留住时光！ ❤️
情感伙伴技能 — 情绪感知、共情回应、基于对话历史的主动关怀。通过 heartbeat 定期检查渠道 session，在合适时机发送自然的问候或关心。
Create or update AgentSkills. Use when designing, structuring, or packaging skills with scripts, references, and assets.
Translate documents, files, or text between languages. Use when the user asks to translate a PDF, document, article, file, or any text content into another language. Trigger phrases include "translate this", "翻译", "翻訳", "번역", "help me translate", "translate to English/Chinese/Japanese/etc."