关于此工具
The same request can sound rude, stiff or perfect depending on wording. This rewriter uses a small instruction-tuned language model that runs on your device to rephrase text in one of four tones: formal and professional, casual and friendly, concise and direct, or polite and tactful. It keeps names, numbers and facts, stays in the language you wrote in, and never sends your draft to a cloud service — useful for work emails, customer replies and messages you would rather keep private. It follows each language’s politeness system: keigo in Japanese, 합니다/해요 in Korean, 您 in Chinese, आप/तुम in Hindi, ครับ/ค่ะ in Thai, vous/tu, Sie/du, usted/tú, Вы/ты and so on.
与 LanguageHelper 上的所有工具一样,AI 语气改写 完全在你的浏览器中运行。文本绝不会被上传、存储或用于训练,因此可放心处理机密文件、客户数据和私人消息。无需注册或安装,加载后即使离线也能继续使用。
首次运行时会下载一个精简的 AI 模型:默认快速模型约 320 MB,最佳质量模型约 480 MB。模型保存在浏览器中,之后几秒即可启动,并可离线使用。模型加载期间会先显示基于规则的即时结果;选择“规则”引擎则完全无需下载。
使用方法
- Choose the target tone and the language of your text.
- Paste the draft you want to rewrite.
- Click “Rewrite” — the result streams in as it is generated.
- Edit the output directly in the box, then copy or download it.
使用场景
- Turn a quick draft into a professional e-mail for a client or manager.
- Make a stiff announcement sound friendlier for a team chat.
- Shorten long messages into concise, direct updates.
- Soften a complaint or request so it stays polite and constructive.
示例
处理前
hey, can u send me the report asap? need it for the meeting tmrw
处理后
Hello, could you please send me the report at your earliest convenience? I need it for tomorrow’s meeting.
处理前
I just wanted to reach out to let you know that the meeting has been moved to Friday.
处理后
The meeting has moved to Friday.
使用技巧
Give the model one message or a few paragraphs at a time — compact models stay more accurate on shorter inputs. Read the result before sending, especially names, dates and numbers, and edit the output box directly if a phrase isn’t quite right. To compare versions, try the same text in two tones, or use Text Diff to see exactly how the wording changed.
常见问题
Which model does it use?
By default the fast model, Gemma 3 270M (about 320 MB); choose “best quality” to use Qwen2.5 0.5B Instruct (about 480 MB). Both are quantized ONNX models executed by ONNX Runtime Web on your device, small enough for a browser and trained on many languages, including all 20 on this site.
How good is it compared with cloud AI?
It is a compact model, so it works best on short texts such as emails and messages. For long documents, rewrite a few paragraphs at a time and review the output.
Will it translate my text?
No. It is instructed to keep the original language. Choose the text language explicitly if auto-detection picks the wrong one.
Is my draft stored?
No. Your text exists only in this browser tab. Closing the tab removes it; only the model files are cached so you don’t download them again.
Does it need a powerful computer?
It uses WebGPU when available and falls back to WebAssembly. Any recent laptop works; on older phones generation is slower but still runs.