Three steps: pick a tool, describe or upload, generate and download. What follows is what makes each of those go well the first time, rather than on the fourth attempt.
Before you start
You need an account and a plan. Sign-in is Google, Apple or an email address, see creating your account. Every paid plan includes all 85+ tools and every AI model, so nothing you're about to pick is gated behind a higher tier. Plans and credits covers the volumes.
1. Pick a tool
Start in the toolbox. It's organized into five groups, image generation, video generation, AI photoshoot, photo editing and audio, and opening a tool takes you into the app with that tool loaded.
Pick by the job, not by the model. Each tool is already wired to the models that suit its work, so choosing the right tool does most of the model selection for you. Choosing the right model covers the rest.
2. Describe or upload
Some tools start from a prompt, some from your own photo, and a few from both.
Writing a prompt. Name the subject, what it's on, what's behind it, the light and the framing. Skip adjectives like "stunning" or "high quality", they don't rule anything out. Writing better prompts goes into it properly.
Uploading a photo. The source image sets the ceiling on the result, and no model recovers detail that was never captured. Good light beats resolution, and the original file beats a screenshot of it. See input image quality.
Nothing prohibited gets through: safety filters run before anything is created. What you can and cannot create is worth two minutes before your first run.
3. Generate and download
Images come back in seconds. Video takes longer, because it's far heavier work per job.
Each generation spends credits, and the amount depends on the model behind it rather than a flat per-image rate. If a generation fails, the credits are returned automatically, usually within minutes. See how credits work.
Download the result and it's yours to use. Work made on a paid plan carries commercial-use rights and needs no attribution.
If the first result isn't right
Change one thing and run it again. Rewriting the whole prompt makes it impossible to tell which change was responsible, and it burns credits learning nothing.
While you're still finding the composition, iterate on a faster model and save the heavier one for the final run, that's the single biggest saving available on credits. And note that a result you simply don't like isn't a failure, so those credits stay spent.
If the generation was stopped rather than disappointing, see why was my generation blocked.
