Docs / Guides
Translation Guide (Pre-Render Translation / Translation Workbench / Image Translation)
The single most important sentence: translation doesn't consume credits. It runs on a completely separate character quota and is available even on the free tier.
The single most important sentence: translation doesn't consume credits. It runs on a completely separate character quota and is available even on the free tier.
Two Entry Points
| Entry | Location | Best for |
|---|---|---|
| Pre-render translation | Homepage input box | Translate while chatting, then read it aloud right after |
| Translation workbench | /studio?mode=translation | Large blocks of text, or where you need to compare source/translation |
Both entries share the same settings (brand, OCR mode) — changing it in one place is enough; no need to configure twice.
Quota: A Separate Ledger From Credits
Translation never deducts credits; it uses an independent, char-counted quota:
| Identity | Daily quota |
|---|---|
| Not logged in | 1,000 chars / day |
| Logged in, no subscription | 5,000 chars / day |
| Subscriber | Monthly quota (see below), no longer subject to the daily limit |
Subscribers' monthly translation quota (counted separately per brand):
| Brand | Plus | Pro | Max |
|---|---|---|---|
| DeepSeek | 500k chars | 1M chars | 2M chars |
| Alibaba Cloud Bailian | 250k chars | 500k chars | 1M chars |
| Volcano Ark | 250k chars | 500k chars | 1M chars |
| Zhipu GLM | 500k chars | 1M chars | 2M chars |
This quota is very generous — Pro's 1M chars is equivalent to translating 30,000+ chars per day, essentially impossible to exhaust. Translation's upstream cost is already extremely low (~¥4 / 1M chars), so instead of folding it into the credit pool we simply grant a large quota.
Other limits:
- Per-request cap: 3,000 chars on the free tier (cost control), relaxed to 20,000 chars for subscribers (a safety cap to stop you pasting an entire book at once). Going over prompts you to send in segments.
- Rate limit: 10 requests/minute (20/minute with your own key)
Four Optional Brands
You can switch the translation provider in Settings:
| Brand | Default | Supports images |
|---|---|---|
| DeepSeek | ✅ Default | ✅ |
| Alibaba Cloud Bailian | ✅ | |
| Volcano Ark | ✅ | |
| Zhipu GLM | ✅ |
All four now support image translation. DeepSeek used to be a pure text model (sending an image errored out directly), but after the official 2026-08-22 release of a vision-capable model, that limitation no longer exists — the default brand can send images, so you don't need to switch brands just to translate an image.
Image Translation
How to Use It
Click the image button in the homepage input box or the translation workbench, pick an image (a screenshot, photo, or chart all work), and the model recognizes the text in the image and translates it.
How the Quota Is Counted
An image counts as a fixed number of characters of quota, regardless of how much text is in it. How many depends on the brand:
| Brand | Per image |
|---|---|
| DeepSeek (default) | 1,500 chars |
| Zhipu GLM | 1,500 chars |
| Alibaba Cloud Bailian / Volcano Ark | 3,000 chars |
DeepSeek dropped from 3,000 to 1,500 on 2026-09-04: derived from its own ¥4 / 1M chars text rate against a real per-image cost of about ¥0.006, 1,500 is the figure that actually matches — the old 3,000 charged double. Zhipu GLM dropped to 1,500 on 2026-09-05 for the same reason — its multimodal tier, GLM-5.3-Flash, is cheaper still. Alibaba Cloud and Volcano Ark now have verified token rates too, but neither publishes a per-image token cap, so a high-resolution image can bill for over a thousand tokens; for them 3,000 may well be under-charging, so it stays put.
On that basis, how many images each tier can translate:
| Identity | DeepSeek / Zhipu GLM (1,500/image) | Alibaba / Volcano Ark (3,000/image) |
|---|---|---|
| Not logged in (1,000 chars/day) | ❌ can't send | ❌ can't send |
| Logged in, no subscription (5,000 chars/day) | ✅ 3 per day | ✅ 1 per day |
| Subscriber (Pro) | ✅ ~666 per month | ~166 per month |
Images Are Never Saved
An image you upload exists only in the current browser session — it isn't uploaded to Soniva's database and doesn't sync with your session to the cloud. After refreshing the page or switching devices, the image preview is gone (the translated text is kept normally).
This is deliberate: images often carry more sensitive info than text (account IDs in screenshots, chats, document watermarks…) — not storing them is the most thorough protection. See data & privacy.
Extra Features of the Translation Workbench
The workbench (/studio?mode=translation) adds a few things beyond the homepage input box:
- Left sidebar: switch translation brand, OCR mode, target language
- Template toolbar: common translation instruction templates, so you don't retype each time
- Quota badge: shows how many chars remain today/this month directly
- Source/translation side-by-side: easier to follow long-text translation
Translate Then Read It Aloud
Translation and reading can be chained: after translating text into a target language on the homepage, simply send it and the current voice will read the translated text aloud.
Remember to switch the voice to a brand/voice matching the target language — reading English translation with a Chinese voice sounds wrong. Microsoft Azure has the broadest language coverage, so try it first for multilingual scenarios.
Related Documents
Was this article helpful?
Anonymous feedback is enabled when you allow analytics and the service is available. You can also report an issue by email.
The email includes this article. You can also copy the details and send them to soniva.cde@gmail.com