Visuals and banners with AILesson 1/4

Lesson 1 of 4 · 10 min

Describing a visual

Subject, style, mood, format: the description that works.

Image generators have made a spectacular leap. ChatGPT, Gemini, Midjourney or Adobe Firefly produce in seconds visuals you once had to commission from a designer. But they all follow the same rule: the quality of the image depends on the quality of the description. This lesson gives you the method to describe exactly what you want, on the first try or close to it.

By the end of this lesson, you will be able to:

  • describe a visual with the five elements that matter;
  • choose the right tool for the visual you need;
  • fix an image in conversation rather than starting over.

The module’s common thread

We will follow Nadia, who is opening a small online tea shop in Sherbrooke (a fictional character, created for this module). She needs a banner for her YouTube channel, thumbnails, Instagram posts and a simple logo. She has no design budget and no design experience.

The five elements of a good description

  1. The subject: what we see, precisely. “A cast-iron teapot on a wooden table”, rather than “tea”.
  2. The style: realistic photo, flat illustration, watercolor, 3D render, vintage style…
  3. The mood: light (soft morning, golden late afternoon), dominant colors, emotion.
  4. The composition: where the subject sits, and above all where to leave empty space to add a title.
  5. The format: square for Instagram, horizontal 16:9 for a thumbnail, very wide for a banner.

Always add what you don’t want: “no text”, “no people”, “no logo”.

Bright, realistic photo of a black cast-iron teapot and two steaming cups on a light wooden table, soft morning light from the left, sage green and cream tones, subject placed on the right with a large empty space on the left for a title, horizontal 16:9 format, no text, no people.

Choosing the right tool

We ran the test: the same teapot description, in ChatGPT and in Gemini.

  • Gemini (Nano Banana models) produces a remarkably realistic photo, with believable steam and depth of field. However, it followed the “large empty space on the left” instruction less closely.
  • ChatGPT respected the composition perfectly, and it excels when the image must contain text (see the next lesson).
Need Suggested tool Why
Image with readable text (thumbnail, poster) ChatGPT The most reliable at writing a title in the image, and it accepts fixes in conversation
Very realistic mood photo Gemini (Nano Banana) Remarkable photographic rendering, free for moderate use
Artistic mood, illustration Midjourney, ChatGPT Very aesthetic results
Commercial use with low rights risk Adobe Firefly Trained on licensed content
Final layout, variations Canva Add text, logo and exact formats

These tools change very quickly: this table reflects November 2026, not a fixed truth. The right habit: test the same request in two tools and keep the best result.

Fixing in conversation

The most common mistake is starting over as soon as the image isn’t right. With ChatGPT, you can fix the image in the same conversation, in one sentence:

  • “Same image, but with late-afternoon light.”
  • “Remove the plant on the right.”
  • “Keep exactly the same scene and add the title ’MORNING TEAS’ on the left, in large white letters.”

Each fix keeps what worked. In three or four rounds, you get the image you want.

The Nadia case

Nadia first wrote “a nice tea picture”. The result: a generic cup on a blurry background, unusable for a banner. With the five elements, she got a bright scene with empty space for her title on the first try. Two fixes in conversation later (“warmer light”, “cast-iron teapot instead of glass”), the image was ready.

Common mistakes

  • A description that’s too short: the tool invents everything else, often blandly.
  • Forgetting space for the title: you can’t add text without hiding the subject.
  • Starting over instead of fixing in conversation.
  • Not saying “no text” when you want to add your own: the tool sometimes adds garbled letters.

Key takeaways

  • Five elements: subject, style, mood, composition, format, plus what you don’t want.
  • Each tool has its strength: ChatGPT for readable text and fixes, Firefly for commercial safety, Canva for finishing.
  • Fix in conversation rather than starting over.
  • Always plan space for your title.

Quiz

1. What are the five elements of a good image description?
2. You want to add your own title on the image. What do you specify?
3. The image is almost perfect, but the light is too cold. What do you do?
4. Which tool should you favor for a thumbnail with a clearly readable title?
5. Why keep your best descriptions in a document?
Next lesson: Banners and thumbnails
Next