What's Changed
新增自定义 LLM 翻译服务，支持 Ollama、LM Studio、OpenAI 等 OpenAI 兼容接口，提供 3 个保存位置，可起名并随时切换
放大镜新增显示选项，颜色格式改为多选，排第一的格式用于复制
剪贴板前台窗口取样频率可在设置中调整
设置窗口与截图工具栏外观更新
修复剪贴板粘贴、钉图文字选择、翻译弹窗、窗口无法拖动调整大小等若干问题
本地模型推荐参数（以 Ollama + Qwen3.5 2B 为例）
- API 地址：
http://localhost:11434/v1（LM Studio 为http://localhost:1234/v1） - API Key：留空
- 模型：点「获取模型」选择
- 附加请求体：
{"reasoning_effort": "none"}，用于关闭思考。实测一句话从约 18 秒降到 0.3 秒；但小模型关闭思考后偶尔会漏行，更看重准确度时可以不填 - 附加要求：例如
Keep product names and key names untranslated，产品名和快捷键会保留原文 - 温度、超时：一般留空；开着思考翻译长文本时，可以把超时调到 120 秒
- JSON 模式：保持开启，服务不支持时会自动关闭后重试
Added a Custom LLM translation service for Ollama, LM Studio, OpenAI and other OpenAI-compatible APIs, with three named slots you can switch between.
New magnifier display options; color formats are now multi-select, and the top one is used for copying.
The clipboard's foreground-window sampling interval is now adjustable in Settings.
Refreshed the look of the Settings window and the screenshot toolbar.
Fixed several issues with clipboard pasting, text selection in pinned images, the translation popup, and windows that could not be resized by dragging their edges.
Recommended settings for local models (Ollama + Qwen3.5 2B as an example)
- API Base URL:
http://localhost:11434/v1(LM Studio:http://localhost:1234/v1) - API Key: leave empty
- Model: click Fetch Models and pick one
- Extra Body:
{"reasoning_effort": "none"}turns thinking off. In testing, a single sentence went from about 18 s to 0.3 s. Small models occasionally drop a line with thinking off, so leave it empty if accuracy matters more. - Instructions: e.g.
Keep product names and key names untranslatedto keep product names and shortcuts in the original language - Temperature, Timeout: usually leave empty. With thinking on, raise Timeout to 120 s for long text.
- JSON Mode: keep it on. If the service doesn't support it, the request is retried without it.
Full Changelog: release-2.0.6...release-2.0.7