↓
Rivya AI Docs

Rivya Reference Files and Upload Limits

Check Rivya image, video, audio, and chat upload requirements, prepare useful references, troubleshoot rejected files, and avoid losing work when changing models.

Last reviewed on September 9, 2026

Choose a model that accepts the files your task needs before preparing an upload. File type, size, count, dimensions, and duration can all matter. Check the model page and its current availability separately: a page may describe inputs for a model that is paused.

Match the File to the Task

What you want to doWhat to check
Use a product photo or another image as a referenceThe image operation, number of references, and each image's role; see image workflows
Animate an image, guide motion, or modify a videoWhether the selected mode accepts images, video, or a combination; see video workflows
Use an existing recordingWhether the specific operation accepts audio; speech generation and new sound generation do not automatically edit recordings; see audio workflows
Ask a chat model about an imageThe model must support image attachments; the current chat file picker accepts JPEG, PNG, and WebP, with a model-specific count

The chat attachment control is not a general PDF, spreadsheet, video, or audio uploader. For an existing conversation, open the correct saved session before adding a new image.

ElevenLabs Audio Isolation is paused, and Rivya currently has no direct available replacement for cleaning an uploaded recording. Suno Sounds creates new sounds; it does not clean your recording. An audio input listed on an unavailable model page is not a runnable feature.

Read the Model's Limits, Not Just the File Extension

These are default upload policies, not a promise that every model accepts these files:

File kindDefault size limit per fileDefault accepted formats
Image10 MBJPEG, PNG, WebP
Video50 MBMP4, MOV / QuickTime, WebM
Audio10 MBMP3, M4A / MP4 audio, WAV, AAC, OGG

Model-specific rules can be stricter or allow more. For example, Nano Banana 2 and Nano Banana Pro allow images up to 30 MB each, while Recraft Remove Background has a 5 MB image limit. Check availability and the selected form before using any of them.

Wan 2.6 illustrates why count and input mode matter: image input uses exactly one JPEG, PNG, or WebP image, at least 256 × 256 pixels and up to 10 MB. Video input uses one to three MP4 or MOV files, up to 10 MB each. Do not mix image and video references in that request. These input rules do not establish current availability.

A file can be below its size limit and still fail a dimension, aspect-ratio, duration, or format check. Rivya checks the actual file format; renaming an extension does not convert a file. Mode-specific and total reference limits still apply even when each file is valid.

Upload and Submit in the Right Order

  1. Confirm the target model is runnable, select its operation, and check the file requirements. Keep original files locally and prepare only what the task needs.

  2. Sign in before uploading. A supported public form can work on the same page after sign-in; Studio is also an option. Follow the login and draft guide to avoid losing unsent text.

  3. Upload through the selected model's form and wait for completion. Check the displayed files, their count, and any role labels. Selecting several files does not mean all were accepted; excess files may not be added.

  4. Write the prompt, check the selected settings and estimated credits, then submit once. An upload finishing does not itself start a generation. The credit guide explains reservation and final settlement.

  5. Review the result against your references. If you change model or mode, check the settings and references again; some may be reset, removed, or no longer valid. Re-upload originals in the new form when needed.

Tell the Model What Each Reference Should Control

Use a clear image for the subject and explain the intended change. If the model supports multiple references, distinguish the subject, style, background, or motion source instead of asking every file to control everything.

For a product-photo edit, you could write:

Use the uploaded bottle as the product reference. Keep its silhouette, cap, and label layout. Replace the background with a pale grey tabletop and add a soft shadow. Do not add text or extra objects.

This is an example instruction, not a tested result or a guarantee of preservation. Check the bottle's proportions, label text, logo, and edges yourself. A reference can guide generation without preserving exact geometry or typography; prepare final brand assets in an editor if accuracy requires it. See choosing a reference-image generator for comparison criteria.

Remove unrelated personal or client information before upload and use files you are entitled to provide. The safe-upload guide covers preparation in more detail.

Fix Rejected or Missing Uploads

ProblemNext step
No upload control, or the control is unavailableCheck sign-in, selected model, mode, and accepted input kinds; changing pages does not add a missing capability
File is too largeExport a smaller copy within that model's limit while preserving any required dimensions or duration
Format is rejectedRe-export in an accepted format; changing the filename extension is not enough
Size is valid but the file is rejectedCheck dimensions, aspect ratio, duration, required media properties, and whether the original file opens correctly
Some selected files are missingCheck per-kind and total counts and allowed combinations; upload only the files needed for this mode
Upload stalls or failsWait for a clear result, confirm the displayed references, and retry missing files; check an already-submitted task before sending it again

If the error persists, follow troubleshooting. Keep the model name, file format and size, error message, and any task identifier. Do not post private files, access links, verification codes, or API keys in a public support request.

Keep Originals and Use the Right API Contract

Use History for saved tasks and download results you need to keep. History is not a permanent archive of every reference or an unsent draft. Keep original files and a local note of important prompts and settings.

For an integration, follow the Files API and the target model API reference. API access has its own key permissions and model-readiness requirements. Where the model requires a model-bound upload, keep the original returned metadata and signed fields; an arbitrary file URL is not a substitute. Re-upload the original if the required metadata is missing or no longer valid.