Choose a Rivya AI model for your task

Start with the result you need and the material you already have. Then use the AI Models catalog to find a model that accepts those inputs and can perform that operation today. A catalog listing is not a promise of access, quality or API support.

Match the operation before comparing models

Write one sentence describing the job. “Animate this approved product image” is more useful than “make something with AI”: it identifies both the input and the intended output.

Video-to-video, motion control and recording cleanup are separate operations. Do not substitute a model merely because it produces the same medium. ElevenLabs Sound Effect V2 and Audio Isolation are currently paused; Suno Sounds creates new sound ideas and does not clean or isolate an existing recording.

For calculations or step-by-step reasoning, AI Calculator and AI Solver provide task-specific chat instructions with a configured model. They are not separate media generators, and their answers still need checking.

Narrow the catalog with search and filters

Use the catalog in this order:

1. Choose the output category: chat, image, video or audio.

2. Choose an input filter if needed: text, image, video, audio or reference material.

3. Search a model name or a short task phrase. Search matches catalog text; it does not test whether a model can complete your request.

4. Open a candidate's detail page to check its current status, supported modes and input requirements.

The result count is the number of matching catalog entries, including entries that may be unavailable. A broad input label does not establish support for every operation involving that input. Credit information helps comparison, but the catalog does not provide a guaranteed cheapest-model ranking or runtime comparison.

If nothing matches, clear the search and filters and try a broader term. Clearing just the search text keeps the category and input filters; the results area's reset control clears all three. You can copy the current page address to preserve the search and filter selection, but the matching entries can change later.

Check the model page and input requirements

Check Model Availability before preparing a run. A paused or coming-soon model cannot accept a new request just because its page remains public; buying credits does not unlock it.

For an available candidate, review the task description, modes, input summary, prompt requirements and FAQ where shown. Then inspect the actual Studio form. Confirm:

• Whether your operation needs a prompt, a file, a URL or several inputs together.

• File kinds, number, size and duration limits, and whether all uploads have finished.

• Available parameters and allowed combinations for the chosen mode.

• The intended role of each reference: subject, composition, style, motion or audio source.

Use Model Fields and Parameters, References and Uploads and Image References for those checks. A reference can guide a result but does not guarantee exact logos, lettering, geometry or product identity. Inspect important details against the original before delivery.

Compare costs using the same intended result

A short clip, a long clip and a chat reply are different units of work. First hold the requested output and essential settings steady; compare the credit estimates each suitable model actually offers. Do not compare a low-resolution draft with a longer, higher-resolution deliverable as if they were equivalent.

Media estimates can depend on the model, settings, input duration or text. Chat costs can depend on tokens and conversation context. The catalog's credit hint is a starting point; check the current form and Credits & Billing for reservation and final settlement behavior.

Choose quality, duration and aspect ratio for the intended placement. A lower setting is useful only if it still tests what matters. More credits, higher resolution or a longer clip do not by themselves fix a wrong instruction or guarantee a better result.

Review one focused attempt before expanding

Try a small version of the real task with a clear acceptance condition. For an approved bottle photo that needs motion, for example, check whether a compatible image-to-video model keeps the bottle shape and label legible while adding the requested movement. This is a suggested evaluation method, not a measured result or model ranking.

Keep the prompt, original files and selected settings. Inspect the result for the condition you defined. If it misses, identify a plausible cause—input quality, instruction, setting or model limitation—and change one relevant factor before another attempt. Each new submission can cost credits; wait for an unclear existing request to resolve before duplicating it.

Changing models may reset settings or remove incompatible references. Recheck the full form after a switch. Save useful results using Output Downloads; history is not a complete project backup. For more examples of choosing by task, read How to Pick the Right Model.

Find the documentation for a specific model

Each catalog detail page is the starting point for that model's description and current access. Use its available mode, input, parameter and FAQ information rather than assuming that a model family's general capabilities apply to every variant.

For a developer integration, also use API Models and API Model Reference. Every /docs/api/models/{id} reference describes that model's API fields and input notices. A listing alone is insufficient: check api_status, supports_direct_generation and api_input_status before choosing a request example. Reference modes may require the Files API even when a text-only mode is callable.

Use the model ID from the API list, not a guessed display name. Studio access and API input support are separate checks. If no currently available model supports the operation and your inputs, there may be no direct Rivya replacement; changing the model name alone will not solve that mismatch.