Browse the help center

Choosing a target or model

Which target to write a prompt for, and which studio model to run it on for photos, images, video and audio, with starting picks and price checks.

Updated Nov 6, 2026

There are two choices to make: which target a prompt is written for in the prompt tools, and which model runs it in the studio. They often match: a Nano Banana prompt runs best on Nano Banana.

Choosing a target

Pick the tool you will actually paste the prompt into. Each target gets a different structure, so a prompt written for one tool can confuse another.

Chat: choose the assistant you use. If yours is not listed, choose Any assistant for a plain structured prompt that works anywhere.
Image: Midjourney expects short phrases and parameters, Nano Banana and FLUX expect full sentences, GPT Image is the one to pick when the image needs exact words in it, and Stable Diffusion gets weighted tags with a negative prompt. If you have not decided, write for Nano Banana and run it in the studio.
Video: Veo prompts can describe sound; Kling and Runway work best with short subject, motion and camera lines; Seedance takes a scene described shot by shot.

Changed your mind after generating? Press Try for to rewrite the same prompt for another target.

Choosing a studio model

Every studio model costs a fixed number of credits per image, per second of video, per job or per 1,000 characters of speech, and all of them draw from the same balance. Match the model to the job and do not pay for more than you need.

Start with the default

On a studio tool, the default model is already chosen for that job, and many tools narrow the list to models that suit it. In the studio, the model list shows only models that support the mode you picked.

Photos of people

For headshots, photoshoots and edits of your own photo, you want a model that keeps the face the same. Nano Banana is the default on most photo tools.

Cheap drafts from text

When you are trying ideas, start on a low-cost image model and move to a stronger one for the final version.

A photo brought to life

For image to video from a photo of a person, Kling 3.0 Standard is the default on the photo-to-video tools.

Video from a prompt, with sound

Veo 3.1 Fast is the default on the text to video and UGC ad tools. Video is priced per second, so the length you choose sets most of the cost.

Voice and music

Use a text to speech model for voiceovers and narration, and a music model for instrumental tracks.

Compare before you spend

The Generate button shows the exact cost for your settings before anything is spent.
Each model page shows its price, inputs, aspect ratios and lengths.
The pricing page shows roughly how many results each plan buys on common models.

If a model fails or its provider's filter rejects the request, see Why was my request blocked. Failed generations are refunded automatically.

Questions

1Is the most expensive model always the best?

No. Credit prices follow what a model costs to run, not how well it fits your job. For simple jobs a cheaper model is often enough, so try it first.

2Why did one model refuse my photo when another accepted it?

Each model provider runs its own safety filter, and some are stricter with photos of real people. Try another model that supports the same mode.

3Can I write a prompt for one tool and use it in another?

Yes, but it works best in the tool it was written for. Use Try for to rewrite it for the other tool in one click.

Still stuck? Contact support

Send us a message and tell us what you tried. Sign in first so we can find your account. What to include is on the contact page.