-
01
How image models think - and how to pick one
≈ 4 minA model turns your words into the image it finds most probable, so vague words produce generic images. Each model has a specialty: speed, text rendering, photorealism - and a published per-image rate.
PracticeOpen the image generator hub and compare the models side by side: rate, strengths, output size.
Compare the image models → -
02
Prompt anatomy: subject, style, light, framing
≈ 5 minStrong prompts name four things: the subject, the visual style, the lighting and the framing. Change one variable at a time and you will learn faster than by rewriting everything.
PracticeGenerate the same subject three times in FLUX 2, changing only the lighting: golden hour, studio softbox, neon night.
Iterate in FLUX 2 → -
03
Text inside the image: posters, menus, UI
≈ 4 minMost models draw letters badly. Two are built for it. GPT Image 2 renders the cleanest typography; Seedream 5.0 plans whole layouts with readable text.
PracticeCreate a poster with a three-word headline and a date in GPT Image 2 - check every letter.
Try GPT Image 2 → -
04
Photorealism and 2K/4K hero shots
≈ 4 minFor campaign visuals, resolution and skin-texture realism decide everything. Nano Banana Pro is the photorealism ceiling of the lineup: 2K by default, 4K on demand.
PracticeRecreate your best lesson-2 image as a 2K hero shot in Nano Banana Pro and compare the detail.
Render in Nano Banana Pro → -
05
Finishing: upscale, clean backgrounds
≈ 5 minGeneration is half the job. Delivery is the rest. Upscale 4x to recover print-ready detail, and cut the background to a transparent PNG for store listings and thumbnails.
PracticeUpscale one image 4x, then remove its background - you now have a production-ready asset pair.
Open the upscaler →
You can now brief, generate and finish images like a pro. The natural next step: keep one face consistent across posts and build a virtual persona.
Build an AI influencer →Finished a course? Take a 20-question test and get a free, verifiable AI certificate.
Take the free test →How to work through this
Each lesson ends in a task you can run in about two minutes. The reading is the shorter half. The part that teaches you something is the run, because the result argues back.
Do the tasks in order. Later lessons assume the settings from earlier ones. Skipping ahead usually means an expensive model on a prompt that was not ready.
Keep the link from every run you liked. It carries the model and the settings, so a good result is repeatable without remembering how you got there.
What the course costs to follow
Reading is free and needs no account. Only the tasks cost credits, and a new account starts with 100 of them without a card.
Free credits cover the tasks that run on fast image models. Tasks on video, music, voice and premium image models need a credit pack. If a lesson suggests an expensive model, it also says what the cheaper draft version is.
Each task shows its credit cost before you run it. A practice run never costs more than you expect.
Questions about
Do I need an account to read this?
No. The lessons are open. An account is needed only to run the tasks, and it starts with 100 free credits.
How long does the course take?
About 22 minutes, tasks included. Reading alone is quicker, but reading without running is how people forget it a week later.
Do I need any experience?
No. The first lesson assumes you have never written a prompt. The later ones assume you did the earlier tasks.
Can I use what I make here?
Yes, on paid packs, under the UGESI terms and the license of the model that produced the result.
