English
Summary
- Corrects model selection for prompt analysis, result evaluation, and variable generation: an explicit evaluation model takes priority, followed by the global optimization model, the caller's fallback model, and the first enabled model.
- Improves reliability when models are deleted or disabled, lists are refreshed, and settings are saved, preventing stale requests from overwriting current state or using unavailable models.
- Includes persisted evaluation, image-recognition, and other known UI settings in backup and restore.
- Includes oversized-image preparation and first-time setup documentation improvements made since the previous release.
Highlights
- Clear the evaluation model setting to restore automatic selection. The UI explains the default order and warns about deleted or disabled function models.
- Earlier evaluation responses no longer interfere with restarted evaluations, and rapid function-model changes are saved in user-action order.
- Image uploads are prepared for compatibility. Material resizing requires confirmation, and cancelling preparation leaves the existing input unchanged.
Product Updates
Desktop
- Receive the shared model-selection, evaluation, and backup fixes.
- Fix repeated quit-time storage saves preventing the desktop process from exiting normally.
- Preserve compatible JPEG inputs and their format while preparing oversized model inputs; PNG conversion uses the PNG output path.
Web
- Improve image upload preparation and feedback in Text-to-Image, Image-to-Image, and Multi-Image workspaces.
- Simplify bilingual first-time model setup, workspace selection, and connection troubleshooting guides, with screenshots from the application.
Core/Infra
- Allow function-model settings to be restored from backups and export persisted known UI settings even when they have not been read during the current session.
- Expand regression coverage for stale requests, model availability, and consecutive settings changes, and replace an obsolete model reference in the migration test fixture.
Fixes
- Fix Basic-mode evaluations preferring the test model over the documented global optimization model, and align related evaluation entry points.
- Skip deleted or disabled evaluation models. Image recognition continues to require an explicitly configured, enabled model.
- Prevent earlier streamed text, results, and errors from overwriting an evaluation that has been cleared and restarted.
- Prevent obsolete model-list responses and refresh failures from overwriting newer state. Function-model callers wait for the latest availability read to finish.
- Notify workspaces after the last text model is deleted, and clear invalid selections after a successful empty-list read.
- Distinguish unloaded, failed, and successfully empty image-model lists so read failures do not clear saved test-column choices.
- Save and reload function-model preferences in operation order, preserving previous settings and displaying an error if a save fails.
- Restore evaluation and image-recognition settings from backups, including persisted known UI settings that have not been read after a restart.
Breaking Changes / Upgrade Notes
- No manual user-data migration is required. Existing custom models and parameter settings are preserved.
- Without an explicit evaluation model, the global optimization model now takes priority. Select a model under Function Models to keep evaluation pinned to it.
- Disabled or deleted function-model preferences remain saved so re-enabling a model can restore the choice. Image recognition requires enabling that model or selecting another available model.
- Development and builds continue to require Node.js 24.
Developer Notes
- The release range is
v2.11.10..HEAD, with 18 content commits before the release metadata commit. - Regression coverage includes evaluation priority, unavailable-model fallback, asynchronous read ordering, consecutive saves, stale evaluation responses, and backup restoration.
- Image upload and documentation improvements include changes already merged into the main release branch after v2.11.10 was packaged.
Installation guide: English | 中文
Source release notes (EN)
中文
概括
- 修复提示词分析、结果评估和变量生成的模型选择顺序：显式设置的评估模型优先，其次使用全局优化模型，再回退到调用方模型和首个启用模型。
- 加强模型删除、停用、列表刷新和设置保存的可靠性，避免旧请求覆盖新状态或继续使用失效模型。
- 完整备份和恢复评估模型、图片识别模型等已保存的界面设置。
- 包含上一版本之后的大图上传自动处理和首次使用文档改进。
亮点
- 清空评估模型设置即可恢复自动选择，界面明确说明默认规则；已删除或停用的功能模型会显示提示。
- 较早的评估响应不再污染重新开始的评估，快速修改功能模型设置时按用户操作顺序保存。
- 大图上传会按兼容要求准备图片；需要明显缩小尺寸时先请求确认，取消处理不会替换现有输入。
产品更新
Desktop
- 同步共享的模型选择、评估和备份修复。
- 修复退出前保存完成后重复进入退出流程、导致桌面进程无法正常结束的问题。
- 为模型图片输入准备保留兼容 JPEG 的原格式，并按尺寸要求处理大图；PNG 使用 PNG 输出路径。
Web
- 改善文生图、图生图和多图工作区的图片上传准备流程及处理提示。
- 简化首次模型配置、工作区选择和连接问题的双语指南，并补充实际界面截图。
Core/Infra
- 将功能模型设置加入备份恢复的允许范围，导出时读取尚未在当前会话访问过的已保存界面设置。
- 加强过期请求、模型可用性和连续设置操作的回归测试，并修正模型迁移测试中失效的型号引用。
修复
- 修复基础模式评估优先使用测试模型、与默认全局优化模型规则不一致的问题，并统一相关评估入口。
- 跳过已删除或停用的评估模型；图片识别仍要求显式配置一个启用的模型。
- 修复清空后重新评估时，旧请求的文本、结果或错误覆盖新评估的问题。
- 修复旧模型列表响应和旧刷新失败覆盖较新状态的问题；功能模型调用会等待最新可用性读取完成。
- 删除最后一个文本模型后仍通知工作区刷新；成功读取到空模型列表时，正确清空失效的模型选择。
- 区分图像模型列表尚未加载、读取失败和成功加载为空，避免读取失败时误清空已保存的测试列选择。
- 按操作顺序保存和重读功能模型设置，保存失败时保留原有设置并显示错误。
- 修复备份遗漏评估模型和图片识别模型设置，以及重启后未读取的已保存界面设置无法导出的问题。
破坏性变更 / 升级说明
- 无需手动迁移用户数据，现有自定义模型和参数配置继续保留。
- 未显式指定评估模型时，默认优先使用全局优化模型；若希望固定使用某个模型，可在“功能模型”中指定。
- 已删除或停用的功能模型设置仍保留，重新启用后可恢复使用；图片识别需要启用对应模型或选择另一个可用模型。
- 本版本沿用 Node.js 24 的开发和构建要求。
开发者说明
- 本次发布范围为
v2.11.10..HEAD，版本元数据提交前共包含 18 个内容提交。 - 回归测试覆盖评估模型优先级、失效模型回退、异步读取顺序、连续保存、旧评估响应和备份恢复。
- 发布说明中的图片上传与文档改进包含此前已经合入主分支、尚未进入 v2.11.10 安装包的改动。
安装文档：English | 中文
仓库版本说明（中文）
macOS note: if macOS reports the app as damaged or cannot verify the developer, this is usually caused by the quarantine attribute on downloaded apps. See the installation guide, or remove it after installing with xattr -rd com.apple.quarantine /Applications/PromptOptimizer.app; for a downloaded DMG, you can run the same command on ~/Downloads/PromptOptimizer-*.dmg before installing.
macOS 备注：如果 macOS 提示“已损坏”或“无法验证开发者”，通常是下载文件的隔离属性导致。请参考安装文档；也可以在安装后执行 xattr -rd com.apple.quarantine /Applications/PromptOptimizer.app，或在安装前对 ~/Downloads/PromptOptimizer-*.dmg 执行同类命令。