The fastest way to waste credits in Rivya is to choose a model because the name feels familiar.
The better move is to decide what the job actually is, what stage the work is in, what the run is starting from, and what you cannot afford to get wrong on the first run. If you want the stricter catalog explanation first, Choosing Models in Rivya is the reference page. This article is the decision layer that sits on top of it.
Start With The Job
Before you compare specific model names, answer the first boundary question:
is this really a chat task?
a still image task?
a video task?
a voice or audio task?
a narrow tool-shaped problem?
That question sounds basic, but it does most of the work. In Rivya, many bad model choices happen before the final model page ever opens. If the output type itself is still unclear, the right first move is often Chat, not a model showdown.
Four Questions First
Most strong model choices in Rivya come from four earlier questions:
what exactly should come out of this run?
am I exploring, controlling, or polishing?
am I starting from prompt only, reference files, uploads, or existing media?
is the first priority speed, finish, or low-risk learning?
Those questions usually narrow the field faster than brand familiarity does.
For example:
a prompt-only exploration run is not the same job as a reference-heavy controlled run
a cheap first-run concept test is not the same job as a premium final asset
a voice-over, a dialogue scene, a planned cleanup task, and a music-first task should not be treated as the same audio decision; cleanup should stop at planning while no cleanup model is available
Match Model To Stage
One of the easiest Rivya mistakes is to keep the same model after the stage of work changed.
The job usually shifts through stages like these:
clarify the brief
explore options quickly
run a more controlled pass
pay for a stronger final finish
That means the right model can change even when the project theme stays the same.
Typical examples:
use chat first when the brief is still unstable
use broader or lower-risk image and video paths when you are still learning
switch into stronger control or higher-finish models once the direction is already proven
choose the audio branch by task shape, not by the vague word "audio"
A model that is right for discovery can be wasteful for the final pass. A model that is right for the final pass can be the wrong place to learn.
Read It As Spending
The most useful fields on a Rivya model page are the ones that tell you whether this run shape actually fits.
The fields that usually matter most are:
strengths
supported modes
reference support or upload shape
direct-generation status
credits hint
sample outputs and FAQ, when available
Read those fields in order: a catalog entry can document a model even when its availability is paused. The current catalog contains 131 documented models, but 86 are available; only an available model with the required execution gate should become a run recommendation.
That is why a famous model name is not enough by itself. If the supported modes, upload shape, or reference limits do not fit the job you have right now, the brand is doing less work than you think.
If you need the shared vocabulary behind those fields, Rivya Glossary and Model Fields and Parameters in Rivya are the best companion pages.
Model-Selection Pattern
The current product usually rewards this pattern:
start at AI Models or the surface hub such as /image, /video, or /audio
open one or two candidates that already fit the output type
compare strengths, supported modes, reference support, and credits hint
confirm the candidate is available, then launch the matching public quick-start block or studio path
switch after the first result if the stage of work changed
That last step matters more than most people expect. Sticking with the previous model just because it already produced something is often how spend drifts.
When To Go Narrower
This page is not the best first stop if:
you need exact field definitions more than decision guidance
uploads and references are the real constraint
you already know the exact workflow and only need a narrower model comparison
you are already down to one family-vs-family question
At that point, the narrower pages are faster:
GPT Image 1.5 vs Midjourney, as a historical comparison while Midjourney is unavailable; use an available image model such as GPT Image 1.5 or Nano Banana 2 for a current run
Where To Go Next
If you need the catalog and field reference, read Choosing Models in Rivya and Model Fields and Parameters in Rivya.
If a candidate is paused, keep the page for comparison but choose an available model with the same required mode; if no true task-equivalent exists, do not pretend a different workflow is a fix.
If uploads and references are the harder constraint, read References and Uploads in Rivya.
If the next question is first-session flow, read How to Run Your First Real Task in Rivya.
If the next question is surface choice, read Image Workflows in Rivya, Video Workflows in Rivya, and Audio Workflows in Rivya.
If you already know the exact kind of work you are doing, the narrower comparison pages are usually faster than staying on a broad selection guide.
Use One Control Brief
When two models both look plausible, compare them with one control brief instead of drifting into separate experiments.
Write down:
the exact task
the starting input
the output format
the must-not-fail constraint
the credit range that still feels reasonable
what would make you switch models after one run
That turns model choice into a controlled decision instead of a popularity contest.
Review Fit Before Switching Models
Before switching, identify the actual failure:
wrong input mode
weak reference handling
poor motion or structure
insufficient finish
too expensive for the stage
brief too vague for any model to answer well
If the failure is in the brief, fix the brief first. If the failure is in the model fit, switch to the model whose strengths match that failure. That discipline is what keeps model exploration from turning into expensive guessing.
