Sticker Studio|表情包设计与制作

SKILL前端设计
前端设计 未标注权限 许可证不明确
安装这个技能

9月份,Sticker Studio 经历了一轮密集的真实生产验证。我们使用实际表情包项目持续测试角色一致性、文字排版、小尺寸显示、夜间模式、视觉占比、局部重做、平台审核以及最终交付流程,并将生产中发现的问题逐步沉淀为可执行规则和回归测试。 这次更新的重点已经不只是“生成更好看的表情包”,而是: 让一套表情包从创意、角色、文案到生产、修改、质检和交付,都能够稳定、可重复地完成。

一句话选型结论

按资料字段整理

Sticker Studio|表情包设计与制作是前端设计类的 Agent Skill,安装方式未标注,权限级别未标注,许可证不明确,有可直接复制的安装命令。技能本质是给模型读的说明书,效果取决于模型与客户端;装完先用一个小任务验证,再放进正式流程。

分类前端设计 安装方式未标注 权限级别未标注权限 许可证许可证不明确
累计安装 未标注 本周 未标注
GitHub 星标 未标注 仓库 未标注
下载量 未标注 技能市场下载数
权限级别 未标注权限 安装前请确认风险范围
许可证 许可证不明确 未说明

安装

未标注
在支持 Skills 的客户端里执行,装完问一句「你有哪些技能」验证
skillhub install user_79ccb752/sticker-studio
  1. 1
    确认客户端支持

    确认你的客户端支持 Skills(Claude、Codex、Cursor 等),并升级到较新的版本。

  2. 2
    拿到技能文件

    执行上面的命令安装;也可以把仓库里的 SKILL.md 手动放进客户端的 skills 目录。

  3. 3
    刷新技能列表

    重新打开客户端,或刷新技能列表,让新放进来的技能被加载。

  4. 4
    验证是否生效

    问一句「你有哪些技能」,确认列表里出现它,再拿一个小任务试跑。

技能档案

前端设计
字段来自技能仓库与公开索引,中文说明由本站标注
作者 / 维护方user_79ccb752仓库 未标注 · 版本 未标注
宿主客户端未标注技能可被这些客户端加载
安装方式未标注安装包 community
权限级别未标注权限安装前请确认风险范围
许可证未说明许可证不明确,商用前请自行确认条款
仓库可用性available技能仓库当前是否可访问
最近核验未标注上游最后核验时间 · 数据更新 1791625142
正文体量17.4 KBSKILL.md 原文大小,越大说明说明书写得越细

典型使用场景

3 类
按分类整理的常见用法,具体以 SKILL.md 为准
出页面与组件

按描述生成页面结构和组件草稿。

调整样式与布局

改间距、配色与响应式布局。

生成设计说明

把设计稿整理成开发能接的说明。

选择建议

装之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有可直接复制的安装命令,不用自己翻文档找入口。

先想清楚的地方

  • 没有标注权限级别,实际能做什么要装好后确认。
  • 许可证不明确,商用或二次分发前先到仓库确认条款。
  • 没有做站内核验,内容以作者仓库为准。
  • 技能只是说明书,实际效果取决于模型与客户端,装完先用小任务试一遍。

更适合

  • 需要在前端设计场景里给模型补一套做法的人。
  • 愿意先读一遍 SKILL.md、确认它会让模型做什么的人。

不太适合

  • 对许可证要求很严、必须明确可商用的人。
  • 只想装完就自动变强、不打算看说明书的人。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
技能权限级别安装方式累计安装宿主客户端安装命令
Sticker Studio|表情包设计与制作当前 未标注权限 未标注 未标注 未标注 有
前端设计 只读 GitHub 仓库 882.3 万 Codex、Claude、Cursor 有
Vercel 组合模式 联网访问 GitHub 仓库 385.9 万 Codex、Claude、Cursor 有
Vercel React native skills 前端 只读 GitHub 仓库 246.9 万 Codex、Claude、Cursor 有
adapt 搜索 浏览器控制 GitHub 仓库 202.7 万 Codex、Claude、Cursor 有
使用我的浏览器 浏览器控制 GitHub 仓库 184.8 万 Codex、Claude、Cursor 有

SKILL.md 全文

17.4 KB
模型实际读到的说明书原文,来自技能仓库

Sticker Studio

P0 Gate — Master Character First

For every CREATE project, the system MUST generate or establish a master character image before batch production. The user must explicitly approve that master character. Record master_character.approved=true and lock the approved asset as master_reference. No user-approved master character = no batch sticker generation. This gate cannot be silently bypassed.

If the user supplies an existing character, treat that asset as the candidate master reference, confirm the visual interpretation, obtain approval, then lock it.

Mission

Turn an idea, copy list, or character reference into a consistent, sendable, platform-ready static sticker pack. The skill acts as IP planner, copy editor, storyboard director, visual-production controller, deterministic image-processing controller, QC system, and delivery packager.

Current public release prioritizes WeChat static sticker packs while keeping platform rules modular.

Entry protocol

  1. Read manifest.yaml.
  2. Classify the user request using rules/intent-routing.md.
  3. Load only the rule/workflow/platform files required by that route.
  4. If a deterministic tool exists for the requested operation, use it instead of image generation.
  5. Maintain project state using schemas/project.schema.json, schemas/assets.schema.json, and schemas/qc.schema.json.
  6. Run QC before packaging and platform delivery.

Route map

  • CREATE → rules/character-lock.md, rules/copywriting.md, workflows/create.yaml
  • COPY → rules/copywriting.md
  • STORYBOARD → rules/storyboard.md
  • GENERATE → rules/generation.md; image generation allowed
  • EDIT → rules/generation.md; semantic image editing allowed, preserve unspecified content
  • PROCESS → rules/pixel-safe.md, workflows/process.yaml, deterministic tools only
  • SPLIT → rules/pixel-safe.md, workflows/split.yaml, tools/cli.py split
  • QC → rules/quality-control.md, tools/cli.py qc-image|qc-project
  • PLATFORM_ASSET → selected platform profile + workflows/wechat-delivery.yaml
  • PACKAGE → platform profile + QC + tools/cli.py package

Non-negotiable principles

  1. Understand before generate. Route intent before choosing a production action.
  2. Preserve beats regenerate. When a usable source image exists, prefer editing/processing it.
  3. Pixel Safe Mode. Resize, crop, pad, alpha, background, format conversion, split, compress, rename and package must not redraw content.
  4. Locked means locked. Do not drift locked character traits without explicit user approval.
  5. Failed-only repair. Repair only failed assets; never regenerate a whole set because one item failed.
  6. Platform rules override defaults. Platform profile > project defaults > generic defaults.
  7. Discussion means no generation. “先讨论” and equivalents block image generation.
  8. No silent creative changes. Do not improve unspecified elements.
  9. Minimum-change interpretation. “其他不变/only change X” means preserve everything except X.
  10. One production error becomes one regression test. Record root cause, rule update and test.

Deterministic execution layer

The following operations MUST use deterministic tooling whenever a usable source file exists:
- inspect metadata
- resize/pad/canvas adaptation
- alpha/transparency verification
- sheet splitting
- file format conversion
- icon technical composition from an approved icon source
- banner technical composition from an approved sticker source
- file/set QC
- packaging

These operations MUST NOT invoke image generation.

Project state

Maintain three machine-readable files in a project workspace:
- project/project.json: project identity, defaults, platform, status
- project/assets.json: canonical source/final asset mapping and per-asset status
- project/qc.json: QC checks, severity, failures, warnings and delivery status

Never infer “第17张” from filesystem order when assets.json exists. Resolve it by asset ID/order in project state.

Character lock

Use templates/character.json and rules/character-lock.md.
- locked: identity-defining traits that cannot drift.
- variable: expression, pose, gesture, direction, props, emotional intensity, composition.
- forbidden: known drift modes and unwanted additions.
- master_references: canonical images/asset IDs when available.

Copy engine

For a user-approved set of up to 24 stickers, seek balanced coverage across high-frequency replies, emotional reactions, social interaction, situations, IP-specific lines and positive/supportive reactions. Score candidates 1–5 on Sendability, Personality and Visuality. Detect semantic/emotional duplication, not only duplicate strings.

Storyboard engine

Before generation, create a production card containing copy, emotion, face, body/action, props, visual contrast/joke, composition, text treatment and output spec. Prefer designs that remain understandable after removing text.

Generation and editing

Use generation only for new visuals or semantic changes such as expression, pose, object, style or character action. For an existing target, preserve all locked traits and all unspecified content.

“Other unchanged” protocol

When the user says “其他不变”, “everything else unchanged”, “only X”, or equivalent:
- set PRESERVE_ALL=true;
- define only requested fields under CHANGE_ONLY;
- preserve all other pixels/semantics to the maximum extent supported by the operation.

Technical changes stay in Pixel Safe Mode. Semantic edits use image editing but must preserve unspecified content.

Platform profiles

Load machine-readable platform settings from platforms/*.yaml. For WeChat use platforms/wechat.yaml. The human-readable description remains in platforms/wechat.md.

If current official requirements conflict with the profile, verify the latest official requirement, update the profile, and add/update platform regression tests before delivery.

QC pipeline

Run file, content, character, small-size and set-level QC according to rules/quality-control.md.
Severity:
- P0 BLOCK: wrong size/format/transparency/count or explicit instruction violation.
- P1 MUST_FIX: wrong text, clipping, neighboring artifacts, severe character drift, unintended elements.
- P2 WARNING: weak sendability, repetitive poses/emotions, low IP personality, weak small-size readability.

Do not package a formal release until all detectable P0/P1 technical failures are resolved or explicitly accepted by the user.

Delivery

For WeChat, execute workflows/wechat-delivery.yaml and validate against platforms/wechat.yaml. Deliver a structured release pack with main stickers, icon, banner, preview, project metadata, QC report and release ZIP.

Regression discipline

Every discovered production error must produce an error record with ERROR_ID, input, expected, actual, root cause, new rule, regression test and status. Re-run python tests/test_runner.py after changes.

V1.1 boundaries

Static stickers only. GIF/video/animation are out of scope. A concept sheet is not a delivery asset. Final stickers are individual files. Image-generation consistency can be evaluated but cannot be guaranteed by deterministic tooling alone.

V1.3–V1.7 Consolidated Production Rules

These rules extend the complete V1.2 executable base. They are additive and MUST NOT be satisfied by silently falling back to older rules.

  • T26: Dynamic text–visual integration (rules/text-composition.md)
  • T27: Single-approved-sticker 750×400 banner (rules/banner-director.md)
  • T28–T30: Copy-driven variation, identity preservation, Minimal Sufficient Variation (rules/character-variation.md)
  • T31: WeChat icon is 100×100 transparent approved-character headshot (rules/icon-director.md)
  • T32: Small-canvas visual priority (rules/small-canvas-composition.md)
  • T33: Dark-mode text readability with adaptive white outline (rules/dark-mode-typography.md)
  • T34: Effective visual occupancy / alpha-bound visual-weight consistency (rules/effective-visual-occupancy.md)
  • T35: Controlled contextual wardrobe variation (rules/contextual-wardrobe-variation.md)
  • T36: Expressive typography + selective keyword emphasis (rules/expressive-typography.md)

Typography execution order

IP typography identity → emotional rhythm → keyword emphasis (NONE/LIGHT/STRONG) → dynamic placement → white outline → light/dark/small-size QC.

Character variation execution order

Expression → Action → Prop → Outfit.
Outfit changes are contextual, never random, and MASTER_REFERENCE identity remains locked.

Final-size execution order

Small-canvas composition → dynamic typography → dark-mode outline → alpha-bounding-box occupancy check → same-size pack visual-weight review → 240×240 export.

V1.8 Wardrobe Library & Selection Director

After MASTER_REFERENCE approval and before batch storyboarding, define/confirm a controlled 3–5 state wardrobe library when clothing can support scene communication. Use rules/wardrobe-library-selection.md, the wardrobe schema and template. Follow Expression → Action → Prop → Outfit State → Outfit Change. Base clothing is default, not mandatory. Never rotate outfits randomly; MASTER_REFERENCE identity overrides wardrobe variation.

V1.9 Series Icon + Market Fit

T38 — Series Icon Differentiation

For same-IP follow-up packs, read rules/series-icon-differentiation.md and maintain series-icon-registry. Same character identity must be preserved, but each series icon must have a clearly distinct signature expression/head state at 100×100 and chat-strip preview size.

T39 — Sticker Market Fit Matrix

Before finalizing a new IP/series, score F/S/P/R/V/E/M/K using rules/market-fit-matrix.md. Use the result to determine whether the concept is primarily traffic-oriented, sendability-oriented, niche/vertical or long-term IP-oriented.

Market Fit is a decision aid, not a reason to override MASTER_REFERENCE, user-approved copy, platform specs or production gates.

V2.0 Adaptive Night-Mode Outline

T40 — Adaptive Night-Mode Outline

All final stickers default to a night-mode safety edge for both text and the main subject. Read rules/adaptive-night-mode-outline.md.

The edge is adaptive, outward-only and anti-aliased. It must be nearly imperceptible on white/light backgrounds while making the subject and text immediately distinguishable on dark/night backgrounds.

Do not use a fixed universal stroke width. Do not create jagged contours, fuzzy halos, dirty fringes or glow effects.

Required QC backgrounds: white, light gray, dark gray, near-black. Run both final 240×240 and realistic smaller chat-preview checks.

When applied to approved artwork, T40 is Pixel Safe technical processing: NO REGENERATION.

V2.1 Maximum Safe Visual Occupancy

T41 — Main sticker + cover + icon

All WeChat production assets default to maximum safe visual occupancy.

Apply T41 to:
- 240×240 main stickers;
- 240×240 cover;
- 100×100 icon.

Minimize meaningless transparent margins. Scale approved art until required content approaches the safe edge, while preserving text, character features, gestures and T40 night-mode outline.

Use visual centering, not merely bounding-box centering. Final QC must include realistic UI-size comparison against representative peer-sized assets.

For approved artwork, scale/reposition only: NO REGENERATION.

V2.2 Platform Review Advisory Preflight

T42 adds advisory platform-review screening before copy lock and before packaging. Known rejection cases trigger warnings and safer alternatives, but the user retains the final creative decision. If the user explicitly keeps the original wording, preserve it and continue unless higher-level safety requirements independently prohibit the request. Consult data/wechat-review-cases.json; never present heuristics as an official blacklist.

V2.3 Character-First Typography Composition

T43 — Character first; typography performs with the character

Typography must not become a detached caption block that forces the IP character to shrink. Start from T41 maximum-safe character occupancy, then adapt text size, line breaks, path and controlled overlap around the pose.

For medium/long copy, reduce/adapt typography before reducing the character. Arc, semi-ring, staggered, directional and controlled-overlap layouts are explicitly allowed. Text may overlap low-information body/clothing areas but must protect eyes, mouth, identity anchors, core gestures and meaning-critical props.

Typography may perform emotion within the sentence through progressive size, spacing, baseline and direction. Standard positive case: 让我缓缓 may use progressively larger characters with looser spacing and a subtly descending baseline to reinforce fatigue.

T43 works with T26 + T36 + T41: text integrates with the visual, acts in the character's voice, and does not steal the canvas from the character.

V2.4 Semantic Reading + Mandatory Occupancy

T44 — Semantic Reading & Local Contrast Typography

Flexible typography must preserve one-glance reading. Keep one semantic sentence as one continuous visual unit by default. Split only when semantic structure/action binding makes the split clearer (e.g. 饼有了,钱呢?, with 钱呢? bound to the open palm). Use a Local Contrast Map and resolve collisions by placement → adaptive text color/value → local separation → subtle backing only as last resort.

T45 — Mandatory Final Occupancy Pass

T41 is now enforced as a required Pixel Safe finalization step for every main sticker, cover and icon. After T40, compute final content bounds, enlarge toward maximum safe occupancy, correct visual center, and compare at realistic WeChat size. Obvious usable transparent margin means the asset is not final. Icon enforcement is stricter: dedicated large-head composition, minimal safe margin, never a small centered portrait.

Rule precedence for final composition

Character identity/safety → semantic readability → character-first maximum occupancy → typography performance → decorative effects.
Typography convenience must never shrink the character or break reading continuity.

V2.5 Production Constitution + Auto-Fit + Clean Delivery

T46 — P0 Production Constitution

Before detailed rules, enforce the eight highest-priority production rules in rules/production-constitution.md: scene-first, identity-first, upper-body default, one-sticker-per-generation, integrated typography, readability-first, six-sample baseline gate, and 240/100/50 A/B/C acceptance.

T47 — Post-Layout Final Auto-Fit

After the full visual (character + copy + outlines + required effects) is complete, run a deterministic Pixel Safe union-bounds auto-fit. Remove meaningless transparent margins, scale proportionally to maximum safe occupancy, correct visual center, then export. Preserve large negative space only when explicitly semantic. Final assets should not require routine manual cropping/zooming during platform upload.

T48 — Clean Delivery Package

The ordinary-user delivery ZIP contains exactly three top-level items: 01_表情包/, 02_源文件/, and 介绍.txt. Internal QC, logs, tests, project JSON, drafts, contact sheets and acceptance reports stay in the Development Complete Package, not the user ZIP.

V2.5 anti-regression precedence

T46 constitution → MASTER gate → six-sample gate → detailed creative rules → T40/T44 contrast/readability → T47 final auto-fit → 240/100/50 A/B/C acceptance → T48 clean delivery. A previously solved issue that returns is a regression, not a new creative choice.

V2.6 Continuity & Repair Constitution

C1–C6 — Stable continuation and minimal-change repair

After T46 P0 Production Constitution and before detailed creative rules, enforce rules/continuity-repair-constitution.md.

Default continuation state is CONTINUE_MASTER, not redesign. Only an explicit request may enter REDESIGN. Local corrections enter LOCAL_REPAIR and must lock all unaffected identity/visual elements.

Hard identity anchors — face, hairstyle, apparent age, body type, primary palette, core outfit identity and art style — may not drift without explicit user direction. Copy-driven expression/action/prop variation remains allowed.

Before production, run four-layer dedup across current and prior same-IP series: literal wording → semantics → send scenario → visual action/composition. Different wording is not enough when the sending reason and visual reaction are effectively duplicated.

Generated scenes must pass Physical Scene Integrity QC: support/contact, occlusion order, intersections, grip, furniture/device placement and limb continuity.

During copy/storyboard planning, nominate primary and backup ICON/COVER/BANNER candidates rather than selecting hero assets only at packaging time.

For repair, use a Minimal-Change Repair Contract: record EDIT_TARGET, lock all unaffected elements, allow only unavoidable dependency changes, and PASS only when the defect is fixed while locked elements remain materially unchanged.

V2.6 execution precedence

T46 P0 Production Constitution → V2.6 Continuity & Repair Constitution → MASTER_REFERENCE lock → six-sample baseline gate → detailed creative rules → T40/T44/T47 final engineering → 240/100/50 A/B/C acceptance → T48 clean delivery.

常见问题

5 问
答案由本站依据条目字段整理
Sticker Studio|表情包设计与制作怎么安装?

在支持 Skills 的客户端里执行:skillhub install user_79ccb752/sticker-studio。安装方式为未标注。装完刷新技能列表或重开客户端。

Sticker Studio|表情包设计与制作支持哪些客户端?

资料没有标注宿主客户端,但只要客户端支持 Skills(Claude、Codex、Cursor 等),一般都能加载。

Sticker Studio|表情包设计与制作权限高吗,安全吗?

权限级别为未标注权限。安装前先看它会读取或写入哪些内容。

Sticker Studio|表情包设计与制作收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于技能本身和所用的模型或客户端,正式使用前按作者说明确认。

Sticker Studio|表情包设计与制作用什么许可证?

许可证标注为 未说明(许可证不明确)。商用或二次分发前请到仓库确认条款原文。

热门专题

按场景挑技能
全部专题 →

技能内容以作者仓库为准;分类、权限级别与许可证状态的中文说明为中文整理,最近核验 未标注。