ChatGPT Image Generation: Posters, Ghibli-Style Art and Thumbnails
Generative AI course · Chapter 24 · Estimated reading time: 10 minutes
Typing "make a Diwali poster" into ChatGPT gives you a pretty picture with diyas, but the shop name is misspelt, the offer is wrong and there is no space for your phone number. An image model draws what you describe, so a vague request gets a generic picture. In this chapter you will learn to write an image brief, the same kind of short note a designer gets, and use it in ChatGPT to make 3 things people actually need: a festival poster for a shop, a Ghibli-style family picture, and a YouTube thumbnail. You will also learn where AI images go wrong and how to fix them.
What you will learn in this chapter
- How text-to-image works, in simple words
- The 6-line image brief
- Where to make images in ChatGPT and the free plan limit
- A Diwali poster with exact text
- A Ghibli-style family picture, with consent and style rules
- A YouTube thumbnail in the right size
- Fix images by chatting, and add text the safe way
- Labels, logos and real people
Related guide
Need a logo, not a poster? Read the guide AI Logo Generator Free (2026): Make a Business Logo and Check It. It fixes AI spelling, checks colour contrast and searches the name on IP India.
Related guide
Making a birthday photo or card? Read the guide Birthday Photo Editing Prompts for ChatGPT and Gemini (2026). It keeps the face real and adds the name and date without spelling mistakes.
1. Suppose the shop needs a poster tonight
Suppose we run Gokul Sweets in Nashik. Diwali is on Sunday, 8 November 2026, and we want a poster for our WhatsApp status and the shop glass: "20% off on dry fruit gift boxes".
We type into ChatGPT:
Make a Diwali poster for my sweet shop
We get a colourful picture of diyas and rangoli. The text says "Happy Diwalli" and "Gokal Sweets". There is no offer, no date, and the picture is square, but our WhatsApp status is vertical.
The image model did exactly what we asked: a Diwali poster for a sweet shop. We just did not tell it what we needed.
2. How text-to-image works
- You describe a picture in words.
- The model draws a new image that matches the words, using patterns it learnt from many images.
- Every time you ask, it draws again, so results change.
The model is very good at look and feel (light, colours, style). It is weaker at exact details: spelling, numbers, logos, hands and the exact count of objects. So your brief should be precise about what matters and you should check those parts every time.
3. The 6-line image brief
The 6-line image brief, top to bottom
| Line | Question | Example |
|---|---|---|
| 1. Use | Where will it be used? | WhatsApp status, shop glass print |
| 2. Size | What shape? | Vertical 9:16, square 1:1, wide 16:9 |
| 3. Subject | What is in the picture? | A gift box of dry fruits with diyas around it |
| 4. Text | Exact words, in quotes | "Gokul Sweets", "20% OFF" |
| 5. Style | How should it look? | Warm festive photo, gold and maroon colours |
| 6. Space and avoid | What to leave empty, what not to add | Empty band at the bottom; no other text, no people |
Copy this template:
Use: [where it will be used]
Size: [9:16 / 1:1 / 16:9]
Subject: [what is in the picture]
Text (spell exactly): "[text 1]", "[text 2]"
Style: [look, colours, light]
Leave space: [where you will add more text later]
Avoid: [anything you do not want]
4. Where to make images in ChatGPT
- Open chatgpt.com or the ChatGPT mobile app.
- Type your brief in a normal chat, or use Create image if you see it.
- To edit an image, stay in the same chat and say what to change.
- Find all your images later in the Images section in the left sidebar (My images).
Free plan: the ChatGPT pricing page says "Limited image creation" on Free. OpenAI does not publish a fixed number, and ChatGPT tells you when you reach the limit and when it resets. So write a good brief first instead of using your free images on guesses.
5. Worked example 1: a Diwali poster with exact text
Prompt:
Use: WhatsApp status and a print for the shop glass.
Size: vertical 9:16.
Subject: an open gift box of cashews, almonds and pistachios on a maroon cloth,
with 5 lit diyas around it and soft marigold petals.
Text (spell exactly): "Gokul Sweets" at the top, "20% OFF on Dry Fruit Gift Boxes"
in the middle, "Till 8 November" below it.
Style: warm festive photo, gold and maroon colours, soft glow.
Leave space: an empty dark band at the bottom for our phone number.
Avoid: any other text, people, logos.
Expected output (yours will be similar, not the same): a vertical, warm poster with the gift box and diyas, the 3 lines of text and a dark empty band at the bottom.
Check every letter: read each text line slowly, letter by letter. Common errors are a missing or doubled letter ("Gokull"), a changed number ("25% OFF") or extra text that you did not ask for.
If one word is wrong:
Keep the image exactly the same. Fix only the text: it must say "Gokul Sweets".
Phone number and address: add them yourself in a photo editor or Canva, in the empty band. A wrong digit in a phone number loses customers, and AI images can still change digits.
Ravindra Bagale's Tip
Shop owners print 200 posters without reading the AI text, and then a customer points out "Diwalli". Image models are still weaker with exact text than with pictures. Keep text short, put it in quotes, read it letter by letter, and add numbers like phone, price and date in an editor where you control every digit.
6. Worked example 2: a Ghibli-style family picture
"Ghibli-style" became one of the biggest AI image trends: photos turned into soft, hand-painted anime scenes. Before you start, 2 rules:
- Consent: use a photo of your own family, and ask everyone in it. Do not upload strangers' or children's photos you do not have the right to use.
- Style, not copying: describe the look (soft watercolour, hand-painted backgrounds), not a specific film's characters or scenes. Keep these pictures for personal use; do not sell them as your own artwork or print them on products to sell. ChatGPT may refuse some requests that copy a specific living artist's style.
Prompt:
Turn this family photo into a soft, hand-painted anime-style illustration:
watercolour textures, warm evening light, gentle colours, a painted sky with
big clouds. Keep the same 4 people, their poses, clothes and positions.
Keep the faces recognisable and friendly. Horizontal 3:2. No text.
Expected output (yours will be similar, not the same): the same family scene redrawn as a warm illustration with painted clouds. The faces are simplified, as in any cartoon style.
Check: 4 people (not 3 or 5), the right person in each place, hands and fingers normal. If someone is missing: "Keep everything, but make sure all 4 people from the photo are in the picture."
7. Worked example 3: a YouTube thumbnail
YouTube's help page recommends a 16:9 thumbnail for videos, as large as possible, with a minimum width of 640 pixels, in JPG or PNG. Uploads from a phone must be under 2 MB (desktop uploads allow more).
Suppose we make a cooking video: "Kaju Katli in 15 Minutes".
Prompt:
Use: YouTube thumbnail for a cooking video.
Size: wide 16:9.
Subject: a close-up plate of diamond-shaped kaju katli with silver varq, on the
right side. On the left, a clean empty area.
Text (spell exactly): "KAJU KATLI" in big bold letters, "15 MIN" in a yellow
circle.
Style: bright, high contrast, sharp food photo, simple background.
Avoid: small text, more than 2 text items, faces.
Expected output (yours will be similar, not the same): a wide, bright image with the sweets on the right and the 2 text items on the left.
Thumbnail checks:
- Shrink it to the size of a phone search result. Can you still read "KAJU KATLI"?
- Is the picture honest? It must show the real dish from your video, not something fancier.
- Export or save it as JPG or PNG; if it is over 2 MB and you upload from a phone, save a smaller copy.
8. Fix images by chatting
Stay in the same chat and change one thing each time:
| Problem | Follow-up |
|---|---|
| Wrong spelling | "Keep the image exactly the same. Fix only the text: '[correct text]'." |
| Too busy | "Keep the subject. Remove the background objects; plain background." |
| Wrong shape | "Make the same image in vertical 9:16." |
| Wrong colours | "Keep everything, change the colours to [colours]." |
| Extra people or objects | "Remove the [object]. Keep everything else." |
If the image gets worse after 3 or 4 fixes, start a new chat with an improved brief.
9. Labels, logos and real people
- Do not use other brands' logos or a famous person's face in your poster or thumbnail.
- Label realistic AI images when you post them. India's IT Rules amendment of 2026 asks platforms to clearly label synthetically generated content; saying "AI-generated" in your caption is the honest habit.
- Do not make images of real people doing things they did not do. That is a deepfake, even as a joke.
- Do not make fake proof like a fake bill, a fake award or a fake news screenshot.
10. Beginner mistakes and quick fixes
| Mistake | What happens | Quick fix |
|---|---|---|
| "Make a poster" with no details | A generic picture | Use the 6-line brief |
| Long text in the image | Spelling errors | Max 2-3 short text lines, in quotes |
| Phone number or price inside the AI image | A wrong digit | Add numbers in an editor |
| Wrong shape for the platform | Cropped poster | Say 9:16, 1:1 or 16:9 |
| Uploading others' photos for Ghibli art | Consent and privacy problems | Your own family, with permission |
| Thumbnail that shows a fancier dish | Viewers feel cheated | Show what is really in the video |
| Many fixes in one message | New mistakes appear | One change per follow-up |
11. Try at home
Try at home
Task 1: a poster
- Pick a real or imaginary shop and an offer.
- Fill the 6-line brief and generate a 9:16 poster.
- Read every letter. Fix one mistake with a follow-up.
Task 2: a thumbnail
- Think of a video you could make.
- Make a 16:9 thumbnail with only 2 text items.
- Shrink it on your phone screen and check that it is readable.
Task 3: style
- Take a photo of a place (your street, a garden), not a person.
- Turn it into a soft watercolour illustration.
- Compare what stayed the same and what changed.
Recap
In short
- An image model draws what you describe; vague requests give generic pictures.
- Use the 6-line brief: use, size, subject, exact text, style, space and avoid.
- ChatGPT's free plan has limited image creation; your images are saved in the Images section.
- Keep text short and in quotes, read it letter by letter, and add numbers in an editor.
- Ghibli-style: your own family photo with consent, describe the look, personal use only.
- YouTube thumbnails: 16:9, minimum 640 pixels wide, JPG or PNG, under 2 MB from a phone.
- Fix by chatting, one change per follow-up; start again after 3-4 failed fixes.
- No other brands' logos, no real people in fake scenes, and label realistic AI images.
Write one 6-line brief today for something you really need, like a status poster or a thumbnail. A clear brief is the difference between a pretty picture and a useful one.
Previous chapter: How AI Photo Editing Works: Profile Photo, Festival Look and 3D Figurine in Gemini (Nano Banana)
Next chapter: AI Product Photos for Amazon, Flipkart and Meesho Sellers (Without Faking the Product)
Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The shop, offer and video on this page are made up for learning. ChatGPT image details were checked on the ChatGPT pricing page and OpenAI's Images help page, and the thumbnail size on YouTube Help, on 19 October 2026; they can change. The Diwali 2026 date was checked against published festival calendars. AI images change every time, so your output will be similar, not the same.