Free AI Tools in Cantonese
粵語
ISO Code: yue
Available Tools in Cantonese
Cantonese AI Capabilities
Free.ai supports Cantonese across 30 tools, powered by state-of-the-art open-source models (Qwen3, Whisper, Madlad). Pick a tool above to start.
- AI chat and text generation in Cantonese (Qwen3 30B MoE)
- Translation to and from Cantonese (Madlad — 419 languages)
- Cantonese speech-to-text transcription (Whisper Large v3)
- Cantonese prompts work for image, video, and music generation
Try AI Tools in Cantonese
No sign up, no cost. Start using AI in Cantonese instantly.
Get Started FreeLanguage Details
| Language | Cantonese |
| Native Name | 粵語 |
| ISO Code | yue |
| Tools Available | 30 |
| Price | Free |
Browse Languages
View All LanguagesFAQ
Cantonese (粵語) has limited but real coverage on Free.ai. AI chat at /chat/, translation at /translate/, and transcription at /transcribe/yue/ all work — these run on multilingual models (GPT-class chat, NLLB / M2M-100 translation, Whisper transcription) that cover the long tail. Text-to-speech voices for Cantonese may be limited or absent; check /voice/yue/ for what is available. Image prompts work better in English. Currently active: chat, search, assistant, write, summarize, paraphrase, grammar, humanizer, detect-ai, expand, simplify, rewrite, email, cover-letter, resume, essay, story, poem, code, coder, translate, image, image/anime, image/logo, image/poster, image/meme, image/wallpaper, video/generate, music/generate, transcribe.
Yes — /chat/ accepts Cantonese prompts and replies in Cantonese. Free signups get 10,000 tokens immediately plus a daily refill pool (resets every 24 hours) that covers normal conversational use. When the daily pool runs out, paid tokens start at $1 = 750,000 tokens; a $5 top-up adds 200,000 tokens at the entry tier. Per-message cost in Cantonese is roughly the same as English on modern multilingual models — no language surcharge.
Partial. The Free.ai UI is auto-translated to Cantonese on /?lang=yue — most strings are covered, but you may see a handful of English fallbacks on newer surfaces until the next translation pass. Hreflang in <head> still emits the yue variant for crawlers. AI tools themselves accept Cantonese input and respond in Cantonese regardless of the UI setting.
TTS coverage for Cantonese is limited. /voice/yue/ falls back to multilingual engines (MeloTTS, premium providers) which may render Cantonese with a generic accent rather than a native voice. If quality matters, translate to a fully-supported language at /translate/ first and synthesize from there, or open a request at contact@free.ai for native-voice priority.
Yes, but expect higher WER. /transcribe/yue/ runs Whisper across all 99 languages it was trained on. For lower-resource languages like Cantonese, WER is meaningfully higher than English — sometimes 25%+ on conversational audio. Pin `/v1/transcribe/?language=yue` so the model does not autodetect wrong. Studio-quality audio helps most.
Yes — /translate/ does both directions. Free.ai uses a mix of NLLB-200 (Meta's 200-language model) and premium models depending on the pair. Popular pairs have dedicated landing pages — try /translate/en-yue/ for English→Cantonese or /translate/yue-en/ for Cantonese→English (where available). The API endpoint is `/v1/translate/?target=yue` — POST your source text in the request body. There is no per-character minimum; short strings and full documents both work.
Yes, with a caveat. /image/generate/ accepts prompts in Cantonese, but every major open and commercial image model (Flux, SDXL, Seedream, Ideogram, etc.) was trained primarily on English-captioned images. Cantonese prompts work, but the tightest prompt-following comes from English. If quality matters, run your prompt through /translate/ first (or use /v1/translate/?target=en in the API) and feed the English version to /image/generate/. Rendering Cantonese text inside images requires Ideogram or a similar text-aware model — most other models will produce gibberish glyphs.
Yes. Cantonese uses the Latin script with locale-specific diacritics, and Free.ai serves UTF-8 end-to-end — chat, transcription output, TTS input, /write/ exports, and PDF/DOCX downloads. Accented characters and locale-specific glyphs render without substitution.
Yes — /write/blog/, /write/essay/, and /write/email/ accept topics in Cantonese and respond in Cantonese. Quality follows the underlying chat model: frontier models (Claude, GPT-class) handle Cantonese well; smaller open-source models may be thinner. Pick a frontier model from the dropdown on /write/blog/ for the best Cantonese long-form output.
Same rates as every other language. $1 buys 750,000 tokens; a $5 top-up gets 200,000 tokens at the entry tier; free signups receive 10,000 tokens immediately plus a daily free pool that covers everyday chat and translation. Cantonese input and output tokenizes at roughly the same rate as English on modern multilingual tokenizers (BPE / SentencePiece), so a Cantonese chat costs about the same as the equivalent English chat. No language carries a surcharge.
Yes. The same endpoints used by the web UI accept Cantonese input transparently — `/v1/chat/` takes a `messages` array in any language and replies in the same one. For translation, POST to `/v1/translate/?target=yue` with the source text. For transcription, pin the language with `/v1/transcribe/?language=yue` so Whisper does not autodetect wrong on short audio. For TTS, POST to `/v1/tts/` with `language=yue` to pick the right voice automatically. Auth: Bearer token from /api/. See /api/ for SDK snippets and rate-limit details.
Free.ai supports 100+ languages with dedicated landing pages at /languages/. The full list covers every language Whisper, NLLB, and the major chat models handle — including all official EU languages, every major Asian language, Arabic and Hebrew, and a growing roster of African and Indigenous languages. Each language page (like this one for Cantonese) lists which surfaces currently support it.