Localize Content in Rivya: Text, Images, Video and Voice

Use Rivya to draft copy for another audience and generate supporting images, video or voice. Treat each output as a draft to review. Changing the website language does not translate an existing asset, and a model's language support is separate from the languages available in Rivya's interface.

This guide covers preparing and checking localized assets. Rivya is not a single control that translates a whole campaign, links every version and publishes it for you.

Start with an approved source and a clear audience

Keep a copy of the source outside Rivya. Specify the target language and region, intended reader, channel and tone. “French for a product page in France” is more useful than “translate into French.” Provide a glossary for product names and terms that must remain consistent.

Separate facts from wording. Features, quantities, prices, eligibility, deadlines and the action you want readers to take must keep their meaning. Sentence order, idioms and examples can change to suit the audience. Formatting a price differently does not authorize converting its currency or inventing a local offer. Write ambiguous dates out in words and retain the correct deadline and time zone.

1. Prepare the source text, approved facts and terms to preserve. Remove information that should not be submitted to a model.

2. In Chat, name the audience and ask for natural wording that preserves those facts. Ask it to flag missing or ambiguous information instead of filling gaps with guesses.

3. Compare the draft with the source. Resolve omissions and changed claims before using the text for images, video or voice.

4. Generate only the assets you need, check each result, and save the approved files with the matching text and version notes.

Rewrite the copy in Chat before generating media

Paste the relevant source into Chat. Do not assume that pasting a page URL makes its full contents available. Image attachments depend on the selected model; the chat attachment flow is not a general PDF, audio or video reader.

For example, the source “Blue ceramic mug, 300 ml. Hand wash only. View the collection.” could become “蓝色陶瓷杯，容量 300 毫升。仅限手洗。查看系列。” for a mainland Chinese product page. This is an illustrative rewrite, not a tested model output. It keeps the material, colour, capacity, care restriction and action without adding dishwasher safety, a discount or shipping terms.

Ask for one usable draft and a separate note about uncertainties. Keep those notes out of the published copy and spoken script. A fluent answer can still omit a restriction or strengthen a claim. Check the meaning against the source rather than approving it because it reads smoothly. See prompt-writing basics.

Create visuals and voice with the right model

Use Image Studio for supported image generation or editing. Attach the references the chosen model accepts and state what must stay unchanged. A reference does not guarantee exact packaging, logos or lettering. Inspect visible text and product details; use a design editor when you need precise typography, layout or final crops.

Use Video Studio to generate supported shots. It is not a timeline editor for replacing captions, trimming an existing film or aligning a new voice track. Prepare translated caption text separately, then check timing, line breaks and placement in a video or subtitle editor. Generated signs and on-screen words also need review.

For narration, put the approved words to be spoken into an available text-to-speech model in Audio Studio. Do not paste “translate this” or editorial notes into the spoken-text field: the model may read those instructions aloud. Voice and language settings vary by model. ElevenLabs Turbo 2.5 forwards its language setting; Multilingual v2 has no separate language-code control in Rivya. Neither the model name nor the interface language guarantees suitable pronunciation for your audience.

Listen for names, numbers, pauses and pronunciation, and check the actual recording length before combining it with video. A translation may take longer to say. Audio cleanup and automatic dubbing should not be assumed available from a general localization workflow; check the specific tool. Review model availability and reference-file requirements before preparing attachments.

Keep versions together in your own files

History helps you revisit submitted tasks, but it is not a localization project manager or a side-by-side source-and-translation review tool. Maintain your own record linking the source, language, approved text, model, relevant settings and downloaded output. A simple filename such as mug-fr-FR-v2 can identify a version without exposing private information.

Copy text drafts and download outputs you need to retain. Do not rely on a browser draft or a provider result link as permanent storage. Saved sessions have their own limits and do not replace a complete asset archive.

Check the estimate before each submission. Separate language versions and retries can create separate tasks and costs; translating a script does not reserve or include its later image, video or voice generation. See credits and billing.

Review the complete version before use

Read the copy in its intended layout, view the visuals and listen to the voice together. Check that facts and qualifications agree across them. Inspect accents and other characters, text direction, line breaks, date formats, units, caption timing and the call to action. A shorter caption must not lose a condition that the source includes.

Have someone who understands the target language and audience review the result when it matters to publication. Model output alone does not establish that a claim is appropriate for a market or that you have the rights to use an asset; see the commercial review checklist.

When the source feature, offer, claim or deadline changes, identify and update every affected version in your own record. Rivya does not automatically synchronize these assets. For a broader campaign example, see the localization workflow for images, video and voice.