dGENdGEN Visual Studio
Small thumbnail sketches on a sheet with one circled, next to a large canvas.

Guide · Working faster

Test ideas without wasting credits

Test with a fast, cheap model, find the idea, then make the final with the best one. A way of working that keeps cost down.

By GENI · 4 min read · updated

Published

Draft cheap, finish good

Most of the work is finding the idea: composition, light, setting. That does not need the most expensive model. Test with a fast one, then make the final with the model that fits the job.

Habits that save credits

  • Make a still before a video.
  • Short clips before long ones.
  • Lower resolution while testing.
  • Change one thing at a time, so you know what helped.

Know the price first

Every model shows its price before you run it, and an assistant sees it too. Set a daily limit for assistants that work on their own.

Models built for drafts

Several models are built for exactly this stage. Each model page says what it is for and shows the price.

  • Nano Banana: cheap first drafts while you find the idea.
  • Recraft 4.1 Flash: cheap drafts, icons and flat illustration in your colours.
  • Z-Image Turbo: high volume with your own LoRAs, at a low price.
  • PixVerse V6: cheap video drafts, with sound when you need it.
  • Vidu: testing image to video ideas cheaply.
  • MiniMax H3: video from cheap 480p drafts to 4K finals, in one family.

A worked test round

You want a hero image of a bakery window at dusk for a website. Round one: four quick drafts on a cheap draft model, each with a different light, to see which mood works. Round two: the winning light, three compositions, still on the draft model. Round three: the best composition once on a stronger model at full resolution.

Eight cheap runs found the idea, and one good run made the final. Starting on the strongest model from the first try would have cost the same eight rounds at the higher price.

Change one thing at a time

Between two test runs, change one thing: the light, the camera or the subject, not all three. When the result gets better you know why, and that knowledge carries over to the next project. Change everything at once and a good result is luck you cannot repeat.

Settings change the price

The price on the button follows your settings. Higher resolution, more images per run, a longer clip and sound switched on all cost more. Turn them down while you test and up for the final.

Watch the number on the button as you change a setting. It updates before you run, so you can see what a choice costs without paying for it.

The Studio start screen, with the line: describe what you have in mind, pick a model, and the price sits right on the button.

Compare once, then commit

Not sure which model suits your idea? The Compare image models template runs one prompt and one source image through six image models side by side. Compare video models does the same for six video models from one start frame.

That is six runs, so use it once to choose a model for a project, not for every test. After that, stay with the winner.

Reuse what worked

Open a result in the viewer and you see the exact prompt it was made with, with a Copy button. When a test works, copy that prompt into the final run instead of rewriting it from memory.

Under a result, Edit keeps the prompt and changes one thing, and As reference starts something new from the picture. Both build on a test you already paid for.

Stills before clips, frames before scenes

A still costs a fraction of a clip. Get the picture right first, then animate it. In a Scenes node, run the frames and check them before the clips. A step that is done never runs again, so a fix later costs only the part that changed.

Habits that waste credits

Each one feels small. Over a project they add up to more than the final runs.

  • Running the same prompt again and hoping. Change something, or stop.
  • Testing at full resolution and with sound on.
  • Making a clip from a still you have not checked at full size.
  • Comparing six models for every new image instead of once per project.
  • Letting an agent run without a daily limit while you try it out.

What a failed run costs

If a run never reached the model, an outage stopped it, or it went wrong on our side, you get every credit back. If the model ran and its safety filter refused your prompt or a reference file, that run is paid for, because the model had already started. Try different wording or another reference.

Questions

Which model is cheapest?

The model button and model pages show the price of each.

Do failed runs cost credits?

The credits and cost guide explains what happens when a run fails.

Is there a limit so I cannot overspend by accident?

Yes. Every account has a daily limit, set in Account under Daily limit, and it counts what an AI agent starts too. Lower it while you test or try a new assistant, so a mistake or a loop cannot empty your balance in one go.

Should the final always use the most expensive model?

No. Use the model built for the job. A cheap model that handles your subject well beats an expensive one that does not. Read what each model page says it is built for, and pick the final model by fit, not by price.

Is Enhance prompt worth the credits?

Often, while testing. It rewrites a short idea into a fuller prompt for a few credits, shown next to the option. If you prefer your own words, Put your own sentence back restores them. Compare the two results once, then decide.

GENIGENI SI OS, the system that runs dGEN Labs. Writes these guides from what the studio can do today.

Read next

Put this guide to work

Open the studio and try what you just read. You start with 1,000 free credits.

Get early access