AI Image and Design

In Midjourney or Ideogram, generate ten images using the subject, style, shot formula from lesson 2. Change only one concrete element between each. You are done when you have ten images and can name the single variable you changed each time.

Generate a portrait of a fictional person, then regenerate it with film grain, visible skin texture, and a single light source added. Save both. You are done when a friend, shown only the second, cannot immediately tell it is AI.

Upscale your best image to at least 3000 pixels wide and have it printed at a local print shop or kiosk as a 6x4 photo. Hold it in your hand. You are done when you can list two things that look worse in print than on screen (usually skin and small text).

Show your best three images to three people without saying they are AI. Write down their first reaction to each. You are done when you have nine reactions noted and can spot which image fooled the most people and why.

Pick one subject (a character, a place, a product) and create a consistent set of four images that clearly belong together in style and color. Log every prompt and seed. You are done when the four sit together as an album without one looking out of place.

Using a character reference, place the same invented person in five completely different settings (beach, office, snow, kitchen, stage). You are done when all five are clearly the same person and the surroundings are genuinely different.

Write a small script that loops a list of twenty prompts through Flux on Fal or Replicate and saves the URLs to a file. You are done when the script runs unattended, all twenty images return, and you know the total cost.

Ask a designer or marketer how they actually use AI images day to day and what still fails them. Write down three concrete pain points. You are done when you can name one workflow gap you could solve with what you have learned.

Create a tiny brand (name, palette, two fonts) and produce three matching assets: a social post, a wide web hero, and a printable poster, all using one style reference and one brand kit. You are done when a stranger can tell the three belong to the same brand.

Construct a ComfyUI text to image graph that includes an upscale step, run it on three prompts, and export the workflow JSON. You are done when a fresh ComfyUI instance can load your JSON and reproduce the same quality.

Build a clean captioned dataset for one subject or style and train a LoRA on it. You are done when the trigger word reliably summons your subject in three new, unseen scenes without overtraining artifacts.

Build an n8n (or scripted) pipeline that reads prompts, generates with Flux, runs a vision model quality gate, brands the pass, and parks results for approval. You are done when ten prompts flow through end to end with only your final approval click.

Write and send one real proposal to a business, or post a finished campaign publicly and ask for feedback. You are done when you have either a reply from a prospect or at least five pieces of honest public feedback on the work.

Take one piece of work you would bill for and check every tool, model, and reference against its commercial terms and any likeness or style concerns. You are done when you have a one page note confirming it is clear to sell, or listing exactly what must change.