AID v2.2.4

AID v2.2.4

中文

• 修复批量视频计划报价对尚未生成提示词的误校验；合并生成在付费文本调用前检查视频模型、凭证和素材。报价不创建任务、不预扣积分。
• 批量视频按项目比例和模型能力使用同一份报价、提交参数，移除历史统一时长残留；预计费用与详情统一显示两位小数，实际计费精度不变。
• 修复视频提示词内部身份与业务编号冲突、场景生图指令干扰及拒答原因解析；保留真实上游用量结算和整批写入保护。
• 修复媒体能力空值、历史数值字符串、Seedance 内部任务参数下发、厂商审核错误和 H3 查询失败收口。
• 完善 GPT-6、GPT-5.6 公开模型目录及缺失价格；修复 Seedance Mini Token 单位、2.0 首帧比例和历史模型池解绑残留。已有有效自定义价格、凭证、倍率和启停状态不覆盖；新模型默认停用。
• 修复本地场景图四宫格拆分、列表刷新、剧集分页、诊断采集超时及共用图片编辑尺寸与蒙版接合。
• 后台恢复正在进行的升级任务终端与进度，完善模型池能力选择、存储地址验证和本地供应商图标。

升级与验证说明

• 本版唯一增量脚本为 sql/v2.2.4.sql，新安装使用 sql/aid-init.sql。不要重新运行已成功的历史脚本；最低直升版本为 1.0.0-beta.1。
• 升级前备份数据库、配置和媒体；迁移及程序切换期间暂停生成写入。MySQL DDL 具有独立提交边界，不承诺整个升级可事务回滚。
• 已核对 MySQL 5.7 新装、上一版直升、完整历史链、重复执行、自定义配置保留、Mini 错误模板及事务故障回滚；三端构建与定向回归按发布记录验收。
• 真实视频生成未对全部供应商逐一测试；未启用供应商和新停用模型不代表账号权限已验证。报价与契约检查不冒充付费生成结果验证。
• 本版不开放自由画布；仅交付已批准的共用修复。

English

• Fix planned batch-video quotes, validate video credentials/assets before paid prompt work, and keep quoting read-only.
• Align batch aspect ratios and submission parameters, remove stale shared durations, and format displayed estimates to two decimals without changing billing precision.
• Correct storyboard prompt identity contracts, refusal handling, media capability validation, provider errors and task-query termination.
• Complete public GPT model catalog and missing pricing; repair Seedance Mini token billing, first-frame aspect ratios and orphaned pool bindings while preserving valid operator settings.
• Fix local scene-image splitting, history refresh, episode pagination, diagnostic latency and shared image-edit geometry; restore active updater progress in the admin UI.

Use sql/v2.2.4.sql for this upgrade and sql/aid-init.sql for new installations. The verified minimum upgrade version is 1.0.0-beta.1. Back up the database, configuration and media, and pause generation writes during migration/switching. DDL is not fully transactional. Newly seeded models remain disabled until operator access checks. Live generation has not been tested with every provider. Free-canvas functionality is not included.

AID v2.2.3

AID v2.2.3

中文更新说明

• 完善 Seedance、WaveSpeed Depth Anything Video、Topaz Proteus、可灵 Omni、MiniMax H3／音乐及腾讯云媒体处理的公共模型目录、能力约束、计费与异步任务链路；新增目录默认停用，管理员需核对自身渠道与价格后启用。
• 腾讯云数据万象和媒体处理支持独立处理 COS，并完善人像分割、音轨分离、智能去字幕、结果转存、媒体类型识别和失败恢复；配置导入使用脱敏预览，保存前执行读写验证。
• 修复模型或能力未单独覆写价格时的保存校验、像素阶梯末档、数组参数编辑、业务默认能力以及媒体输出规格回显问题。
• 修复手动场景“仅保存”后提示词丢失、媒体任务取消、图片多结果持久化以及官方更新清单 CDN 不可用时的验签回退。
• Web 增加媒体批量停止的逐项状态，并明确已执行任务可能继续完成。
• 自由画布仍处于未完成阶段，不在本次公开发布范围。
• Topaz 实际生成受当前供应商账号额度限制，仅完成可执行的协议、报价、校验、退款与配套构建检查；Seedance 2.5 与可灵 Omni 的真实输出还受当前上游服务开通状态限制，不宣称已完成成功成片验证。
• 数据库升级执行 sql/v2.2.3.sql；脚本可重复执行，并保留管理员已保存的密钥、启停状态、价格和业务选择。升级前请备份数据库、配置与用户媒体，并在维护窗口暂停媒体任务写入。

English Release Notes

• Expanded the public model catalog, capability constraints, pricing, and asynchronous task handling for Seedance, WaveSpeed Depth Anything Video, Topaz Proteus, Kling Omni, MiniMax H3/music, and Tencent Cloud media processing. New catalog entries stay disabled until administrators verify their own provider access and pricing.
• Added a dedicated processing COS path for Tencent Cloud CI/MPS and improved portrait segmentation, audio-track separation, subtitle removal, result persistence, media-type detection, and recovery. Configuration import now uses a redacted preview and validates read/write access before activation.
• Fixed model-save validation for inherited prices, open-ended pixel tiers, array parameters, default business capabilities, and media-output specification editing.
• Fixed manual-scene prompt persistence, media-task cancellation, multi-image result persistence, and signed fallback when the official CDN update manifest is unavailable.
• Added per-item batch-stop results in the Web client and clarified that already-running tasks may still finish.
• The unfinished free-canvas feature remains excluded from this public release.
• Live Topaz generation was limited by the current provider account quota; protocol, quotation, validation, refund, and build checks were completed where executable. Seedance 2.5 and Kling Omni live output also remain limited by current upstream service enablement, so this release does not claim successful output generation for those paths.
• Apply sql/v2.2.3.sql when upgrading. The migration is repeatable and preserves administrator-managed credentials, enablement, prices, and business selections. Back up the database, configuration, and user media first, and pause media writes during migration.

AID v2.2.2

AID v2.2.2

中文更新说明

• 修复 TokenDance MiniMax H3 多参考视频错误回落到单首帧装配、从而拒绝多张参考图的问题；参考图会按模型能力去重、连续编号并在超过上限时于提交前拒绝。
• 增加默认关闭的腾讯云数据万象图像主体检测公共能力，可返回主体名称、0～100 置信度及原图像素坐标；只接受系统已登记的 COS 对象或受控 HTTPS 图片，不缓存识别结果、不创建媒体任务、不扣减用户积分。
• 后台增加主体检测配置和真实联通测试，可配置凭证来源、超时、限流与 COS 取图方式，并展示实际供应商调用次数、主体数、耗时、状态和请求 ID。
• Web 修复关闭剧本历史侧栏后退场遮罩仍短暂拦截点击的问题，并移除角色、道具形态标题中不可编辑的序号前缀。
• 自由画布仍处于未完成阶段，不在本次公开发布范围。
• 数据库升级执行 sql/v2.2.2.sql；该脚本只补齐主体检测默认配置且不覆盖管理员已保存值。升级前请备份数据库、配置与用户媒体，并在维护窗口暂停媒体任务写入。

English Release Notes

• Fixed TokenDance MiniMax H3 multi-reference storyboard requests falling back to single-frame assembly and rejecting multiple references. References are deduplicated, renumbered consistently, and rejected before submission only when the configured limit is exceeded.
• Added an opt-in Tencent Cloud CI object-detection capability that returns object labels, 0–100 confidence values, and source-pixel boxes. It accepts only registered COS objects or controlled HTTPS images, stores no detection cache, creates no media task, and consumes no user credits.
• Added admin configuration and live connectivity testing for credential source, timeouts, rate limits, and COS image access mode, including provider-call count, object count, elapsed time, status, and request ID.
• Fixed the story-history drawer mask briefly blocking the page after close, and removed non-editable numeric prefixes from character and prop form titles.
• The unfinished free-canvas feature remains excluded from this public release.
• Apply sql/v2.2.2.sql when upgrading. The migration only adds missing object-detection defaults and preserves existing administrator values. Back up the database, configuration, and user media first, and pause media writes during migration.

AID v2.2.1

AID v2.2.1

中文更新说明

• 错误处理页将本地错误、已提交报告及自愿附带的排查资料统一保留 7 天，并区分报告发送状态与官方人工处理进度。
• 诊断报告改为逐条提交并强制绑定唯一请求；服务器只读排查资料必须先完成连接验证，避免重复报告或资料静默遗漏。
• 图片模型新增结构化图层分离输出，按顺序保存底图、透明图层及尺寸、层级、位置等元数据；Seedream 5.0 Pro 可通过独立能力启用该官方模式。
• 图片计费支持按每张实际输出的像素档位结算，预冻结按最大输出估算并只退不补；无法读取、格式或透明通道不合法的图层结果会失败退款，不会留下已扣费的半成品。
• 后台补充图层分离能力和逐张像素档位计费编辑，并改进诊断授权、连接验证、提交进度及人工处理状态展示。
• Web 用户端本次没有代码变化；自由画布不在本次发布范围。
• 数据库升级执行 sql/v2.2.1.sql。已在隔离 MySQL 5.7 库验证新安装、从 v2.2.0 升级、从最早公开版本按顺序升级，以及本版 SQL 重复执行；升级前仍需备份数据库、配置与用户媒体，并在维护窗口暂停媒体任务写入。

English Release Notes

• Extended local error, submitted report, and optional troubleshooting-data retention to seven days, and separated delivery state from official handling progress.
• Bound each diagnostic submission to exactly one request and required successful connection verification before attaching read-only server troubleshooting data.
• Added structured image-layer outputs with ordered background and transparent layers plus dimensions, stacking order, and placement metadata. Seedream 5.0 Pro can enable its official layer-decomposition mode independently.
• Added per-output pixel-tier settlement. Pre-authorization uses the maximum estimate and settlement only refunds; unreadable or invalid layer results fail and refund instead of leaving charged partial output.
• Added admin controls for layer decomposition and per-image pixel tiers, with clearer authorization, connection-verification, submission, and handling-status feedback.
• The Web client is unchanged in this release; the unfinished free-canvas feature is not included.
• Apply sql/v2.2.1.sql when upgrading. Isolated MySQL 5.7 tests covered fresh installation, upgrade from v2.2.0, the full ordered chain from the earliest public version, and repeat execution. Back up data, configuration, and user media first, and pause media writes during migration.

AID v2.2.0

AID v2.2.0

中文更新说明

• 新增默认关闭的错误诊断与日志报告发送。服务端、后台和 Web 可关联同一请求及异步任务，保留脱敏后的供应商错误与浏览器证据；管理员在服务器出口 IP 获授权后，可将报告加密送达官方，查看异步发送结果并人工重试未确认报告。不会自动上传全部服务器日志，也不会重试原业务生成请求。
• 修复并发请求和定时任务可能引起的注解缓存死锁；改进供应商欠费与鉴权错误的区分。
• 扩充图片编辑模型能力与业务配置，支持蒙版选区、扩图、打光、输出尺寸和多结果幂等保存；能力模板只补齐缺项，新增能力默认停用，不更改既有报价。
• 模型可选配图片输入 URL 代理模板；未启用时保持原有资源签名方式。视频参考模式支持参考音色 ID，模型接口向 Web 返回当前业务的提示词字符上限。
• 后台优化模型池候选与参数提示、视频能力设置和长列表横向浏览；Web 编辑弹窗按模型能力校验提示词长度，项目详情及诊断状态请求避免重复提交。
• 修正仍使用旧版错误默认值的交流二维码资源路径；管理员自定义地址保持不变。
• 数据库升级执行 sql/v2.2.0.sql。已在隔离 MySQL 5.7 库验证从最早公开的 v1.0.0-beta.1 按顺序升级，以及本版 SQL 重复执行；升级前仍需备份数据库、配置与用户媒体，并在维护窗口暂停媒体任务写入。

English Release Notes

• Added opt-in error diagnostics and encrypted report delivery. Server, admin, and Web evidence is correlated by request and task. After the server's outbound IP is authorized, administrators can send a sanitized report, inspect asynchronous delivery, and manually retry an unconfirmed report. This does not upload all server logs or retry the original generation request.
• Fixed a possible annotation-cache deadlock under concurrent requests and scheduled tasks, and distinguished provider credit exhaustion from authentication failures.
• Expanded image-edit model capabilities and business configuration for masks, expansion, relighting, output dimensions, and idempotent multi-result storage. New capabilities remain disabled by default; existing pricing is preserved.
• Added optional image-input URL proxy templates, reference voice IDs for video scenes, and model-specific prompt limits exposed to Web editors.
• Improved model-pool validation, video capability settings, and horizontal navigation in long admin tables. Web editors enforce the selected model's prompt limit and coalesce duplicate project-detail and diagnostic requests.
• Corrected only the legacy default community QR-code path; administrator-defined URLs remain unchanged.
• Apply sql/v2.2.0.sql when upgrading. An isolated MySQL 5.7 test covered the ordered chain from v1.0.0-beta.1 and repeat execution. Back up data, configuration, and user media first, and pause media writes during migration.

AID v2.1.9

AID v2.1.9

中文更新说明

• 新增 Topaz Labs 图片增强接入与停用模型目录，支持按输出像素计费；启用前需由站长配置密钥、核对价格并实测。
• 新增 New API 站点接入、可选普通用户访问令牌、文本和图片模型目录与真实生成测试；未验证执行协议的媒体模型仍禁止导入。
• 供应商管理增加三方聚合／官方厂商分类、排序和更清晰的模型编辑；图片 SKU 可配置像素计费并调整优先级。
• 通用文本、图片、视频任务支持入库后异步提交，文本任务显示估算进度；图片处理支持原像素宫格和矩形裁切。
• 修复分镜任务防重、剧本对话在模型停用后的恢复和停止交互，并改进请求参数错误及 Agnes 只读连通性测试。
• 内置供应商图标、默认头像和验证码背景随程序提供，已有自定义图片优先；补齐已导入 TokenDance MiniMax H3 的未自定义场景。
• 数据库升级使用 sql/v2.1.9.sql，MySQL 5.7 隔离库已验证从最早公开版本连续升级与重复执行。升级前请备份数据库和配置；本版不覆盖已有密钥或自定义报价。

English Release Notes

• Added a disabled-by-default Topaz Labs image enhancement catalog with output-pixel billing. Administrators must configure credentials, review pricing, and verify generation before enabling models.
• Added New API site integration, optional user access tokens, text and image catalogs, and real generation checks. Media models without a verified execution protocol remain blocked from import.
• Improved provider grouping and ordering, model editing, pixel-based image SKU pricing, and SKU priority controls.
• Added durable asynchronous submission for text, image, and video tasks, estimated progress for text tasks, and original-pixel grid and rectangle cropping.
• Fixed storyboard task deduplication and screenplay chat recovery after a model is disabled; improved validation errors and Agnes read-only connection checks.
• Bundled default provider icons, avatars, and captcha backgrounds while preserving custom images. Completed verified MiniMax H3 scenes on existing TokenDance catalog models when not customized.
• Apply sql/v2.1.9.sql when upgrading. The migration was tested on MySQL 5.7 from the earliest public baseline and on repeat execution. Back up data and configuration first; existing credentials and custom prices are preserved.

AID v2.1.8

AID v2.1.8

中文更新说明

模型与账号

• 已删除模型保留历史编码与任务关联；管理员可以重新添加同编码的活动模型，两个活动模型仍不能重名。历史别名不会把新模型的调用误导向旧模型。
• 后台逻辑删除账号后释放手机号和邮箱，保留订单及流水的用户 ID 关联。历史已删除账号的联系方式由幂等迁移释放；后台新增账号也遵守已有的注销后再次注册间隔。

分镜视频

• 图生视频无需额外上传参考图即可使用已选分镜主图；多参视频允许只填写提示词。已上传的图片、视频、音频及模型能力限制保持不变，报价与提交使用一致的素材口径。

安装与升级

• 修复安装脚本和源码构建器仍访问旧源码仓库的问题。Docker 与非 Docker 模式均从统一公开仓获取构建器和固定版本源码，网络故障时不替换正在运行的部署。
• 既有环境按 SemVer 顺序执行 sql/v2.1.8.sql；新装基线已包含对应结构。迁移兼容 MySQL 5.7、可重复执行，不修改模型密钥、价格、启停状态、订单或流水。升级前请备份数据库、部署配置和用户媒体。

English Release Notes

Models and Accounts

• Deleted models retain historical codes and task associations. Administrators can create a new active model with the same code, while duplicate active codes remain prohibited. Legacy aliases no longer redirect the new model to an older one.
• Soft-deleting an account releases its phone number and email without removing orders or ledger links. A repeatable migration handles previously deleted accounts, and administrator-created accounts respect the configured re-registration interval.

Storyboard Video

• Image-to-video generation can use the selected storyboard image without an additional upload. Multi-reference video generation can start from a text prompt alone. Supplied image, video, and audio references and model capability limits remain in effect; quotes and submissions use the same reference inputs.

Installation and Upgrades

• Fixed installer and source-builder references to the retired source repository. Docker and non-Docker installs now fetch the builder and version-pinned source from the unified public repository; download failures preserve the running deployment.
• Existing installations apply sql/v2.1.8.sql in SemVer order; fresh installations include the matching schema. The MySQL 5.7-compatible migration is repeatable and does not change model credentials, prices, enablement, orders, or ledger entries. Back up the database, deployment configuration, and user media before upgrading.

AID v2.1.7

AID v2.1.7

中文更新说明

视频模型调用配置

• 修改模型的真实上游标识时，仍继承原模型标识的能力协议绑定会同步更新，避免后续视频生成继续请求旧模型。
• 管理员明确配置的独立上游 Endpoint ID 不会被改写；新建任务使用更新后的配置，已创建任务的路由快照不变。

数据库与升级

• 已有环境按顺序执行 sql/v2.1.7.sql。脚本只修复仍指向平台模型编码的旧绑定，兼容 MySQL 5.7、可重复执行；不改密钥、价格、启停状态或自定义 Endpoint ID。
• 新安装初始化数据已使用真实上游标识，无需额外初始化变更。经隔离库验证，最低一键直升版本仍为 v1.0.0-beta.1。升级前请备份数据库、部署配置和用户媒体。
• 本版仅涉及服务端；后台管理端和 Web 用户端没有新的业务代码变更。

English Release Notes

Video Model Invocation Configuration

• Changing a model's real upstream identifier now updates capability protocol bindings that still inherit the previous model identifier, preventing subsequent video jobs from calling the old model.
• Explicitly configured upstream Endpoint IDs are preserved. New tasks use the updated configuration; routing snapshots of existing tasks remain unchanged.

Database and Upgrades

• Existing installations apply sql/v2.1.7.sql in version order. The MySQL 5.7-compatible, repeatable migration repairs only bindings still pointing to the platform model code. Credentials, pricing, enablement, and custom Endpoint IDs are untouched.
• Fresh-install data already uses real upstream identifiers, so no additional initialization change is needed. Isolated database validation keeps the minimum one-click upgrade version at v1.0.0-beta.1. Back up the database, deployment configuration, and user media before upgrading.
• This release changes the server only. The admin and Web applications have no new business-code changes.

AID v2.1.6

AID v2.1.6

中文更新说明

批量生成与媒体任务

• 修复专业版多参数视频提示词在有效场景下遗漏项目全局风格的问题；空输出、缺少场景或时长仍按无效结果处理。
• 媒体任务只截取安全长度的展示摘要，避免大批量提示词在入库阶段超过 TEXT 容量；完整请求与实际模型输入不变。
• Web 批量视频部分失败时展示任务返回的安全失败原因，成功产物照常回填；普通生成、恢复和续生使用一致的提示。
• 配音试听在供应商明确拒绝时返回可操作的安全提示，不再误报为缺少异步任务 ID。

数据库与升级

• sql/v2.1.6.sql 条件式修正已有库的菜单名称及早期媒体任务提示词摘要列字符集为 utf8mb4。新安装基线也使用相同字符集；迁移不覆盖已有数据或自定义配置，历史乱码仍须依据原始来源单独恢复。
• 已在隔离 MySQL 5.7.20 实例从 v1.0.0-beta.1 初始化库顺序执行全部正式增量至本版，整条链重复执行两遍均成功。本版一键直升最低版本继续为 v1.0.0-beta.1。升级前请备份数据库、部署配置、证书及用户媒体。

开源发布与文档

• 统一公开仓地址为 aid-studio，更新安装、升级、Release、模型目录和社区入口；既有自定义配置不会被覆盖。
• 优化中英文 README 的产品介绍与流程画布展示，增加英文 Docker、systemd、备份、HTTPS 和外部中间件部署指南；新增 Web 静态生成检查。
• 本版包含服务端与 Web 用户端修复；后台管理端没有业务代码变化。

English Release Notes

Batch Generation and Media Tasks

• Fixed missing project-wide style in valid professional multi-parameter video prompts. Empty output, missing scenes, and missing duration still fail validation.
• Bounded the stored display summary of media-task prompts to avoid TEXT column overflow during large batches. Full requests and actual model inputs are unchanged.
• The Web client now shows safe item-level reasons for partial video-batch failures while retaining successful outputs. Initial runs, recovery, and continuation use consistent messaging.
• Voice preview now returns an actionable safe message when a provider explicitly rejects a request instead of reporting a missing asynchronous task ID.

Database and Upgrades

• sql/v2.1.6.sql conditionally converts existing menu-name and legacy media-task prompt-summary columns to utf8mb4. The fresh-install baseline matches. Existing data and custom settings are preserved; previously corrupted text requires separate recovery from an authoritative source.
• The full migration chain from the v1.0.0-beta.1 baseline through this release passed twice on an isolated MySQL 5.7.20 instance. The minimum supported one-click upgrade version remains v1.0.0-beta.1. Back up the database, deployment configuration, certificates, and user media before upgrading.

Open-source Release and Documentation

• Unified public repository URLs under aid-studio across installation, upgrades, Releases, model catalogs, and community links without overwriting existing custom configuration.
• Refined the bilingual README and flow-canvas presentation, added English Docker/systemd and backup/HTTPS/external-service guidance, and added a Web static-generation check.
• This release includes server and Web fixes; the admin application's business code is unchanged.

AID v2.1.5

AID v2.1.5

中文更新说明

模型、计费与任务安全

• GPT-5.6 的展示名称更新为 GPT-5.6 Sol，平台稳定编码继续使用 gpt-5.6，真实上游模型统一为 gpt-5.6-sol，既有 Skill 与业务引用无需修改。
• GPT-5.6 Sol 的内置计费基线更新为不超过 272K 与超过 272K 两档，覆盖输入、缓存读取、缓存写入、可见输出和推理 Token；迁移只修正已知旧内置价格，不覆盖站长自定义 SKU。
• 文本模型失败结算统一依据真实调用证据：请求未发送，或上游明确最终拒绝且没有产生用量时全额解冻；已返回用量按实际用量结算；超时、断链、私有状态码、5xx 或结果不确定时继续保守结算。
• OpenAI 兼容（含 DeepSeek）、Gemini 与 TokenDance 文本协议会在预冻结和排队提交前校验密钥、地址及端点，避免本地配置错误进入可能计费边界。

Web 用户端

• 新增账号安全设置中心，并优化账号设置表单的间距、操作按钮和信息层级。
• 修复素材提示词编辑、登录反馈、生成后提示词保留、弹窗提示词格式回显及确认弹窗布局问题。
• 修复无需生成分镜图的创作模式仍会打开分镜图编辑弹窗的问题；跨步骤返回时会按当前创作模式决定是否打开编辑器。

分镜生成与任务恢复

• 修复包括 Agnes 在内的视频模型及分镜图片批量任务在部分失败后续生时只统计本轮镜头的问题；现在会按原批次完整镜头范围汇总历史成功与本轮结果，避免产物已全部落库但父任务仍无法成功收口并被超时回收。
• 批量生成部分失败时保留每一项的安全错误提示，客户端优先展示 failedItems[].userMessage，缺失时再回退到任务级 errorMessage。

数据库与升级说明

• 已有环境按迁移顺序执行 sql/v2.1.5.sql。脚本兼容 MySQL 5.7、可重复执行，并为早期安装链补齐 GPT-5.6 Sol 缺失的结构化能力、协议绑定和模型别名。
• 已在 MySQL 5.7.44 隔离库从 v1.0.0-beta.1 连续执行全部正式增量到 v2.1.5，整条迁移链重复执行两遍均成功；当前初始化基线要求的表、字段、索引、公共配置键和公开模型缺失数均为 0。
• 自定义模型价格、协议路由、启停状态、密钥、网关、倍率与模型池不会被本次迁移覆盖。正式渠道最低一键直升版本继续为 v1.0.0-beta.1。
• 本版本包含服务端和 Web 用户端变更；后台管理端代码没有变化。升级前仍请备份数据库、部署配置、证书和用户媒体目录。

开源文档

• README 鸣谢新增 Linux.do 社区，并明确交流群仅用于开源交流与技术支持，不含广告及隐形消费。
• 重构 README 中英双语入口和产品首屏，补齐英文部署、首次创作、维护指南、贡献指南、安全说明、支持说明、社区准则与 Issue/PR 模板。
• 增加服务端与后台管理端 GitHub Actions 自动构建门禁；安装示例优先使用正式渠道，第三方归属统一收录于 NOTICE，并保留标准 MIT 许可。

English Release Notes

Models, Billing, and Task Safety

• GPT-5.6 is now displayed as GPT-5.6 Sol. The stable platform code remains gpt-5.6, while the upstream model is normalized to gpt-5.6-sol, so existing Skills and business references remain compatible.
• The built-in GPT-5.6 Sol billing baseline now has separate tiers for contexts up to 272K and above 272K, covering input, cache reads, cache writes, visible output, and reasoning tokens. The migration updates only recognized legacy built-in prices and preserves administrator-defined SKUs.
• Text failure settlement now follows provider-independent call evidence: requests that were never sent, or explicit final refusals with no usage, are fully released; reported usage is settled as reported; timeouts, disconnects, private status codes, 5xx responses, and uncertain outcomes remain conservatively settled.
• OpenAI-compatible providers (including DeepSeek), Gemini, and TokenDance validate credentials, endpoints, and base URLs before reservation and again before queued submission, preventing local configuration errors from entering a potentially billable boundary.

Web Client

• Added an account security settings center and refined spacing, action hierarchy, and visual structure in account settings.
• Fixed media prompt editing, login feedback, prompt retention after generation, prompt formatting in dialogs, and confirmation-dialog layout.
• Fixed image-editor dialogs opening in creation modes that do not generate storyboard images. Cross-step navigation now respects the active creation mode.

Storyboard Generation and Task Recovery

• Fixed resumed batch storyboard image and video tasks, including Agnes-backed video jobs, counting only the current retry instead of the complete original shot range. Historical successes and current results now converge into one authoritative terminal summary.
• Partial batch failures preserve item-level safe error messages. Clients should prefer failedItems[].userMessage and fall back to the task-level errorMessage only when needed.

Database and Upgrade Notes

• Existing installations apply sql/v2.1.5.sql in migration order. The script is MySQL 5.7 compatible and idempotent, and it restores missing structured capability, protocol binding, and model alias records for GPT-5.6 Sol on early installation chains.
• The complete versioned migration chain from v1.0.0-beta.1 through v2.1.5 was executed twice on an isolated MySQL 5.7.44 instance. No required current-baseline tables, columns, indexes, public configuration keys, or public models were missing afterward.
• Custom model pricing, protocol routes, enablement, credentials, gateways, multipliers, and model-pool assignments are preserved. The stable one-click upgrade floor remains v1.0.0-beta.1.
• This release changes the server and Web client; the admin application source is unchanged. Back up the database, deployment configuration, certificates, and user media before upgrading.

Open-source Documentation

• Added the Linux.do community to README acknowledgements and clarified that the community group is dedicated to open-source discussion and technical support, without advertising or hidden charges.
• Reworked the bilingual README entry and product hero, and added English deployment, first-creation, maintenance, contribution, security, support, community, Issue, and pull-request guidance.
• Added GitHub Actions build gates for the backend and admin application. Installation examples now prefer the stable channel, third-party attribution is centralized in NOTICE, and the standard MIT license is retained.