↓
Rivya AI Docs

Image References in Rivya: Choose, Attach, and Review

Choose and attach image references in Rivya. Check the selected model's file limits, explain each image's role, and compare the result with the original.

Last reviewed on November 11, 2026

Use a model and operation that accept image uploads. A reference can guide a product's appearance, composition, or style, but does not guarantee an exact copy. The currently available Chat Studio models accept text only; they cannot inspect an attached image. For image generation controls, see Image Studio.

Choose the image operation before uploading

Use Seedream 5.5 Text to Image for a new image from text, or Image to Image for reference-led edits. Each submitted reference adds to the edit budget.

Layer Decomposition is a separate operation: its output limit includes the base image, and the reserve covers all potential outputs. Keep every returned layer and its JSON layout details when continuing in an editor. Rivya Image Upscale enhances an existing image; its price tier depends on verified output pixels, not the scale factor alone. Check the selected page for formats, size limits and output settings before uploading.

Assign each reference a purpose

PurposeWhat to explainWhat to check afterward
Product appearanceWhich object's shape, color, and label should remain recognizableCompare those details with the real product.
CompositionSubject position, crop, and space for textCheck the layout at the intended aspect ratio.
Style and lightingPalette, light direction, or atmosphere to borrowCheck that the treatment has not changed product facts.

Product, composition, and style are ways to explain a brief. They do not create separate input slots or automatic controls. An image can serve more than one purpose; state the priority.

Prepare and attach the images

  1. Choose a model in the model directory, check availability, and confirm that the operation accepts image uploads. A model that generates images from text may have no image-upload input. Sign in before uploading or submitting.

  2. Check the selected model's upload requirements, including count, file format, size, and dimensions. The examples below apply only to those image-to-image operations; other models and modes have their own rules. File sizes use MiB (binary megabytes). Use only images you have the right to submit, and remove unnecessary private information; see safe uploads.

  3. Wait for upload completion and check the attached previews. Selecting more files than the remaining allowance can leave only a subset attached. Uploading sends the files beyond your browser before you generate. Removing a preview removes it from the current form; it does not confirm deletion elsewhere.

  4. If the operation accepts a prompt, identify each image by its visible content and explain what to use from it. After changing the model or mode, recheck the attachments and settings: incompatible or excess references can be removed and settings can reset.

Image-to-image exampleReference imagesFormats and upload size
Flux 2 ProUp to 8 imagesJPEG, PNG, or WebP, up to 10 MiB per image
Nano Banana 2Up to 10 imagesJPEG, PNG, or WebP, up to 30 MiB per image

Resolve conflicting instructions before generating

Start with the smallest set that explains the task. Add another view only if it contributes a needed detail and the model supports the count. More references do not guarantee better fidelity.

For a model accepting two references, a brief might say: “Keep the bottle's appearance from the product photo. Use only the off-center placement from the second image, leaving space on the left for a heading. Do not copy its product or label.” This is an instruction example, not a tested result or an assurance of exact reproduction.

If one image suggests a blue bottle and another suggests a green one, state the product priority or remove the conflicting reference. Keep the necessary constraints even in a short prompt; the prompt-writing guide helps make them explicit.

Review against the original, not just the previous output

Compare product appearance, layout, and style separately. Inspect lettering, logos, small packaging details, hands, reflections, and edges at the intended display size. A convincing image can still contain incorrect product facts; use the commercial review checklist before delivery.

Before using a generated image as a new reference, compare it with the original product photo. Otherwise, an error in one output can carry into later versions. References do not guarantee exact geometry, repeated character details, or consistency throughout a series.

Before another attempt, identify the mismatch and decide whether to clarify priorities, reduce conflicting inputs, or select a more suitable model. Check credit billing and the estimate before submitting: a new generation request can have a separate credit cost, and some models charge for reference inputs.

Save the files you will use again

Find submitted tasks in History and download approved results before using them as new references. Keep original inputs and the accepted prompt, model, and settings yourself. Saved sessions do not provide a complete editable archive or permanent file access.

For a coordinated set, continue with image workflows. The reference-image generation guide and brand consistency guide discuss specific applications of these checks.