↓
Rivya AI Docs

Rivya Video References: Images, Frames, and Motion

Choose image, frame, video, and audio references for supported Rivya video modes. Check file requirements and compare the complete result with your original sources.

Last reviewed on November 11, 2026

Choose references for the selected video model and mode. A product image, an opening frame, and a motion video have different roles; supplying one does not enable another input or guarantee that the result preserves every detail. For the full submission process, see Video Studio; for planning several shots, see video workflows.

Separate generation, editing and upscaling

Gemini Omni 1.2 Flash has distinct reference, first/last-frame and video-edit modes. Keep frame input separate from references or video. Editing follows the server-verified source duration and does not send the generation duration field.

Grok Imagine Video 1.5 Preview now includes a priced 1080p option while retaining its one-image input limit. Rivya Video Upscale processes an existing clip within a selected output box; it does not turn a text prompt into a new scene. Review source duration and the current reserve before submitting either video operation.

Match the material to a supported input

MaterialWhat to confirm before uploading
Product or scene imageThe model accepts images in this mode; describe which appearance or composition details matter
First and last framesCheck how the selected mode assigns opening and closing images; general references do not become endpoints just because a prompt calls them frames
Motion video with a subject imageUse a model offering this combination; distinguish the subject’s appearance from the source movement
Existing video for another operationCheck the operation, its companion inputs, and required video properties; a video input does not provide a full editing timeline or preserve every cut and sound
Speech or audio referenceCheck whether the operation accepts audio and requires an accompanying image or video; switching on generated audio does not attach a recording

Check the model directory and availability guide. A public model page is not evidence that every reference type is currently usable. Text-only modes do not gain image control because a prompt mentions a photo.

Kling 2.6 Motion Control and Kling 3.0 Motion Control require one image and one video. The source video must be at least 3 seconds long, with a maximum of 10 seconds for image orientation or 30 seconds for video orientation. For these models, Rivya's upload limits are 10 MiB per JPEG or PNG image and 95 MiB per MP4 or MOV video; MiB means binary megabytes. Check the selected model's other input requirements too: the two versions have different rules for image and video dimensions. These requirements do not apply to other video models.

Prepare references with one clear purpose

Choose material that shows the details you need to preserve. For a bottle, the original product photo should show its shape, color and label clearly. A background reference may suggest composition or lighting; it should not silently replace the product with a different bottle.

If the image needs preparation, use a suitable tool in Image Studio or an image editor. Compare the prepared image with the original before using it for video. An edited or generated reference can already contain errors.

Use only material you have the right to upload, and remove private information the task does not need. Follow the safe upload guidelines. Check the selected operation's file count, formats, size, dimensions, and duration; frame rate and bitrate also matter where required. For the upload checks, see references and uploads. For speech inputs, also read audio uploads.

Attach and test in four steps

  1. Sign in, select a compatible model and choose the required mode. Decide what each reference should contribute before attaching it.

  2. Upload to the correct fields and wait for completion. Check the files and their roles, including any first/last-frame fields and required image, video, or audio pairs. Selecting files on your device does not confirm that they all reached the form or request.

  3. Describe the movement and the details to retain. Choose the available duration, ratio, resolution, and audio settings. Check the attachments and credit estimate, then submit once.

  4. Watch and listen to the whole result. Compare it with the original references and save a useful version before another attempt.

For a video mode that accepts an image, a focused instruction could be:

Use the attached bottle photo for the bottle’s appearance. Keep it still on a plain desk while the camera moves slowly closer. Keep the whole bottle and label visible, retain its color, and add no objects.

This is an instruction example, not a tested output or an exact-match guarantee. Use quality, duration and aspect ratio to check actual settings. A prompt cannot enable an unsupported control.

After changing the model or mode, recheck the settings and every attachment, including special reference fields. A change can reset parameters or remove incompatible or excess files. Uploading sends files beyond your browser before generation; removing a preview from the form does not confirm deletion elsewhere.

Review what changed and resolve conflicts

Inspect product shape, color, label, scene layout, camera movement and sound timing throughout the clip. Look for changes that appear after the opening, such as a distorted label or an added object. Natural-looking motion does not establish that the real product can perform the action shown.

If the wrong detail was copied, clarify the reference’s role or replace the conflicting material. Keep the required product facts fixed; do not relax them simply to obtain smoother motion. When multiple references are supported, state their priorities and remove unnecessary contradictions rather than adding more files by default. More references do not guarantee greater fidelity.

Change one identifiable part of the request and record the difference in the result. A new generation request can have a separate credit cost; some models' estimates depend on reference inputs or their duration. Review credits and billing and failed tasks and refunds. A technically successful result that does not meet your creative expectations is not automatically refundable.

Before commercial use, follow the review checklist. Compare with the original product, not only the previous generated clip. Add exact logos, interface text or subtitles in editing where needed. Reference inputs do not replace a full editing timeline or guarantee consistency between separately generated shots.

Keep the source material with the accepted result

Use History to find tasks, and download approved outputs before reusing them. Keep your own originals, accepted prompts, model names and settings alongside the files. Saved sessions do not guarantee a complete editable project or permanent access to every reference and output.

If sound needs separate production, use an appropriate audio workflow and combine files in an editor. A video operation that accepts audio may use it for a specific task, such as lip sync; it does not provide a general mixing timeline. For a product-specific brief, continue with video prompts for product demos.