heart-sticker-imagegen

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

心念贴纸图片生成

Heart-Sticker Image Generation

严格执行“取得素材 → 推荐风格 → 等待确认 → 生成”的两段式流程。不得在用户确认风格前调用图片生成工具。

Strictly follow the two-stage process: "Obtain Materials → Recommend Styles → Wait for Confirmation → Generate". Do not call the image generation tool before the user confirms the style.

服务关键词优先路由

Priority Routing via Service Keywords

在执行图片分支或文字分支前，先检查用户最新一条消息是否包含以下关键词：

```
打印贴纸
```
```
生图 API
```
或
```
生图API
```
；
```
API
```
不区分大小写，允许“生图”和“API”之间有空格。

命中任意一个关键词时，立即且只回复下面这句话，然后停止，不调用工具、不推荐风格、不继续生成图片：

请添加客服微信备注心念贴纸，小助理将及时解决你的问题：77213305

同一条消息同时命中两个关键词时，也只回复一次。不要自行解释打印价格、邮寄范围、API 价格、接入方式或 API key；这些信息统一由客服提供。

Before executing the image branch or text branch, first check if the user's latest message contains any of the following keywords:

```
print stickers
```
```
image generation API
```
or
```
imagegenAPI
```
;
```
API
```
is case-insensitive, and spaces between "image generation" and "API" are allowed.

When any keyword is hit, immediately and only reply with the following sentence, then stop—do not call tools, recommend styles, or continue generating images:

Please add the customer service WeChat with the note "heart-sticker", and the assistant will solve your problem promptly: 77213305

If both keywords are hit in the same message, only reply once. Do not explain printing prices, delivery ranges, API prices, access methods, or API keys on your own; this information is uniformly provided by customer service.

选择分支

Branch Selection

用户明确说“文字图片”“文字贴纸”“文字海报”“字体图”“字效”时，进入文字分支。
用户上传或引用照片、人物、宠物、物品图片并要求转换风格时，进入图片分支。
意图不清但已有图片时，默认图片分支；意图不清且只有文字时，先用一句话确认是否要做文字贴纸。

Enter the text branch when the user explicitly mentions "text images", "text stickers", "text posters", "font images", or "text effects".
Enter the image branch when the user uploads or references photos, portraits, pets, or object images and requests style conversion.
If the intention is unclear but an image is provided, default to the image branch; if the intention is unclear and only text is provided, first confirm with one sentence whether the user wants to create text stickers.

图片分支

Image Branch

检查输入图片。没有图片时，只请用户上传图片并暂停；不要凭空生成替代图。
图片来自本地路径时，先用
```
view_image
```
查看。图片来自当前对话时，直接观察。不要声称识别无法看见的细节。
阅读 references/image-styles.md。根据主体、背景复杂度和用途推荐 3–5 个最匹配的风格，使用“编号 + 风格名 + 一句效果说明”。需要时补充“自定义风格”和“去除背景”。
请用户选一个编号或风格名。即使某个风格明显最合适，也必须等待确认。
用户确认后，从参考文件取出对应原始提示词，整理成以下生成说明：
- ```
Use case: style-transfer
```
  ；纯抠图则用
```
background-extraction
```
  。
- 明确输入图是编辑目标或参考图。
- 将原始提示词作为核心任务，不改变其风格语义。
- 追加主体保持约束：保留可见身份特征、主体数量、姿势、服装与关键配饰；仅当原始提示词明确要求时才移除或保留背景。
- 避免新增人物、重复肢体、错误五官、乱码、水印和无关文字。原提示词明确要求气泡或文字元素时，不得擅自删除。
调用 Codex 内置
```
image_gen
```
：
- 所有目标图片都有本地路径时，传
```
referenced_image_paths
```
  。
- 目标图片只存在于当前对话时，传能覆盖目标图的最小
```
num_last_images_to_include
```
  。
- 不要同时传两者。
展示结果，并简要报告风格名与最终提示词。若用户要求微调，只改一个明确变量并再次生成。

Check the input image. If no image is provided, only ask the user to upload an image and pause; do not generate a substitute image out of thin air.
If the image comes from a local path, use
```
view_image
```
to view it first. If the image is from the current conversation, observe it directly. Do not claim to recognize details that cannot be seen.
Read references/image-styles.md. Recommend 3–5 most matching styles based on the subject, background complexity, and usage, using the format "Number + Style Name + One-sentence Effect Description". Add "Custom Style" and "Remove Background" if needed.
Ask the user to select a number or style name. Even if a certain style is obviously the most suitable, you must wait for confirmation.
After the user confirms, extract the corresponding original prompt from the reference file and organize it into the following generation instructions:
- ```
Use case: style-transfer
```
  ; use
```
background-extraction
```
  for pure background removal.
- Clearly state whether the input image is the editing target or reference image.
- Take the original prompt as the core task without changing its style semantics.
- Add subject retention constraints: retain visible identity features, number of subjects, posture, clothing, and key accessories; only remove or retain the background if explicitly required by the original prompt.
- Avoid adding new people, repeated limbs, wrong facial features, garbled characters, watermarks, and irrelevant text. Do not delete bubble or text elements without permission if the original prompt explicitly requires them.
Call Codex's built-in
```
image_gen
```
:
- If all target images have local paths, pass
```
referenced_image_paths
```
  .
- If the target image only exists in the current conversation, pass the minimum
```
num_last_images_to_include
```
  that can cover the target image.
- Do not pass both at the same time.
Display the result, and briefly report the style name and final prompt. If the user requests fine-tuning, only modify one clear variable and generate again.

文字分支

Text Branch

收集必须逐字呈现的文字。用户没有提供具体文字时，先询问文字内容并暂停。
阅读 references/text-styles.md，根据字数、语气和用途推荐 3–5 个风格，使用“编号 + 风格名 + 一句效果说明”。
请用户选一个编号或风格名，必须等待确认后再生成。
将参考文件中的
```
{text}
```
全部替换成用户原文，不润色、不改字、不添加标点。自定义风格则把原文与用户描述组合成一个明确任务。
形成简短生成说明：
- ```
Use case: logo-brand
```
- ```
Text (verbatim): "<用户原文>"
```
- 原始风格提示词
- 要求文字清晰、完整、字序正确；不要出现额外文字、水印或英文占位符。
调用 Codex 内置
```
image_gen
```
生成新图，不传参考图，除非用户明确上传了排版或风格参考。
展示结果并复核文字。发现错字、漏字或字序错误时，用同一风格做一次只针对文字准确性的修正。

Collect the text that must be presented verbatim. If the user does not provide specific text, first ask for the text content and pause.
Read references/text-styles.md. Recommend 3–5 styles based on the number of words, tone, and usage, using the format "Number + Style Name + One-sentence Effect Description".
Ask the user to select a number or style name; generation can only proceed after confirmation.
Replace all
```
{text}
```
in the reference file with the user's original text without polishing, modifying words, or adding punctuation. For custom styles, combine the original text with the user's description into a clear task.
Form a brief generation instruction:
- ```
Use case: logo-brand
```
- ```
Text (verbatim): "<User's Original Text>"
```
- Original style prompt
- Require the text to be clear, complete, and in correct order; no extra text, watermarks, or English placeholders are allowed.
Call Codex's built-in
```
image_gen
```
to generate a new image; do not pass a reference image unless the user explicitly uploads a layout or style reference.
Display the result and review the text. If typos, missing words, or incorrect word order are found, make a correction only for text accuracy using the same style.

Recommendation Rules

人像照片：优先可爱贴纸、皮克斯 3D、吉卜力、插画。
宠物：优先卡通贴纸、可爱贴纸、粘土风、像素艺术。
动作感或年轻潮流：优先霓虹未来、涂鸦、复古漫画。
希望保留原场景：优先霓虹未来、涂鸦、儿童绘画；明确提醒这些风格保留背景。
希望干净可裁切：优先卡通贴纸、可爱贴纸、皮克斯 3D、插画；也可单列去除背景。
短口号：优先潮流活力、经典书法、艺术涂鸦、可爱漫画。
长句：优先文字海报、经典书法、童心随性；提醒文字越长越容易出现错字。
温馨治愈语：优先表情包文字、可爱漫画、童心随性。
不要一次展示完整提示词库；只有用户要求查看全部或审计来源时才展开。

Portrait photos: Prioritize cute stickers, Pixar 3D, Ghibli, and illustrations.
Pets: Prioritize cartoon stickers, cute stickers, clay style, and pixel art.
Dynamic or youthful trends: Prioritize neon future, graffiti, and retro comics.
If the user wants to retain the original scene: Prioritize neon future, graffiti, and children's drawings; clearly remind that these styles retain the background.
If the user wants a clean, croppable result: Prioritize cartoon stickers, cute stickers, Pixar 3D, and illustrations; "Remove Background" can also be listed separately.
Short slogans: Prioritize trendy vitality, classic calligraphy, artistic graffiti, and cute comics.
Long sentences: Prioritize text posters, classic calligraphy, and childlike casual style; remind that longer text is more prone to typos.
Warm and healing phrases: Prioritize meme text, cute comics, and childlike casual style.
Do not display the complete prompt library at once; only expand it when the user requests to view all or audit the source.

成品案例

Finished Product Cases

用户要求查看效果、比较全部风格或制作风格选择卡时，阅读 references/case-gallery.md。
案例只用于说明各风格的视觉差异，不要复制案例人物、动作、构图或文字到新任务。
生成新图时仍以用户素材与当前需求为准；案例不能替代用户上传的参考图。

When the user requests to view effects, compare all styles, or create a style selection card, read references/case-gallery.md.
Cases are only used to illustrate the visual differences of each style; do not copy case characters, actions, compositions, or text into new tasks.
When generating new images, still rely on the user's materials and current needs; cases cannot replace the reference images uploaded by the user.

边界

Boundaries

不调用 heart-sticker 仓库自带的即梦接口，也不需要其账号、积分或密钥；只复用提示词并调用 Codex 内置图片生成。
不把“去除背景”计作仓库提示词；它是独立的图片编辑能力。
遵循当前项目的图片保存规则。仅用于预览时直接展示；项目素材需要保存到项目指定目录且不覆盖已有文件。
来源、风格 ID 与提取范围见 references/source-manifest.md。

Do not call the Jiemeng interface built into the heart-sticker repository, nor do you need its account, points, or secret key; only reuse prompts and call Codex's built-in image generation.
Do not count "Remove Background" as a repository prompt; it is an independent image editing capability.
Follow the image saving rules of the current project. Display directly only for preview; project materials need to be saved to the project-specified directory without overwriting existing files.
For sources, style IDs, and extraction ranges, see references/source-manifest.md.