--- title: Best way to learn AI image generation is by trying date: "2025-04-01T04:06:09Z" lastmod: "2025-04-01T04:06:23Z" categories: - llms - education wp_id: 4007 description: "The fastest way to understand AI image generation is not theory but playful experimentation across replacements, stickers, interfaces, products, and environments." keywords: [AI image generation, experimentation, prompting, creative practice, ChatGPT images, learning by doing] --- ![Best way to learn AI image generation is by trying](/blog/assets/3ArgIxR.webp) I figured I should spend a few hours on the native image generation bandwagon and push the bounds of my imagination. Here are some of my experiments with image generation on ChatGPT. - **Replacements**: [Replace the person with this image](https://freeimage.host/i/3ArgIxR) (after uploading a photo of [Naveen](https://www.linkedin.com/in/naveengattu/)) - **Sticker**: [Create a transparent comic-style sticker of a lady chef featuring this person happily cooking salad](https://iili.io/3ASLUJ9.png) (after uploading a photo of my wife) - **Meme sticker**: [Create a transparent sticker of a Vadivelu meme](https://iili.io/3ASs9YF.png) - **Meme**: [Create an image of Vadivelu looking up from a well. No caption. Make it look like a frame from a Tamil film.](https://iili.io/3AUwYcQ.png) - **Recipe**: [Invent a vegetarian dish that has NEVER been created. Describe the ingredients and procedure first. Then draw a mouth-watering image of the dish](https://chatgpt.com/share/67ea44a2-1034-800c-997a-b43eafbd59f1). ([Another version](https://chatgpt.com/share/67ea4457-427c-800c-b429-aacc1b96bd0b)) - **Infographics**: [Create a detailed comic infographic explaining the double slit experiment.](https://iili.io/3AUeRPs.png) - **Slides**. [Draw a beautiful infographic highlighting these 6 accessibility testing aspects, with apt icons and visuals](https://chatgpt.com/share/67ea4f76-4710-800c-89a1-addf54d6ec12). - **UI mockups**. [Draw the screenshot of a chat application incorporating these features](https://chatgpt.com/share/67ea5497-8bbc-800c-887f-9ad365ee10cb): ... - **Product ideation**. [Draw an iSuit designed by Apple and Iris van Herpen. Show multiple views showcasing all features. Then write a product description](https://chatgpt.com/share/67ea4671-18e0-800c-8adc-621c46ddca48). - **Interior design**. [Draw a biophilic office where the ceiling is a mirrored hydroponic garden, reflecting lush greenery downward to create the illusion of working in a floating forest](https://chatgpt.com/share/67ea49c8-73c8-800c-bf83-dad8a68852a2). - **Meeting room design**. [Draw a modern office with sound-absorbing ‘whisper walls’ covered in fractal patterns that visually dampen noise pollution while doubling as collaborative whiteboards](https://chatgpt.com/share/67ea497b-ed90-800c-be02-0291fa36f410). - **Restaurant design**. [Draw a marble dining table with a river flowing through it, serving conveyor belt sushi as the dishes float gently on the water on top of plates](https://chatgpt.com/share/67ea4b6c-f584-800c-9392-814ba1bad5a3). - [A sentient toaster with googly eyes, riding a unicycle through a library](https://chatgpt.com/share/67ea51a1-4754-800c-ac8d-2442649a3922). - [A painting painting itself, but it's struggling with existential dread](https://chatgpt.com/share/67ea51bd-e1dc-800c-8f90-74b5545e2f01). - [Photo of a gym where people work out by lifting their own regrets](https://chatgpt.com/share/67ea529a-0bcc-800c-80d0-ff2f158c107c). **Here's what I learnt.** - The refusal rate is low, but it does refuse to generate **some** copyrighted material like Calvin & Hobbes strips. - Using a prompt to generate the description and using THAT to prompt for images helps. - A more imaginative model (like [DeepSeek](https://chat.deepseek.com/), maybe [Grok](https://x.com/i/grok)) can help create good prompts that ChatGPT can execute faithfully. - There are hallucinations that experts can detect. E.g. Naveen's and Vadivelu's faces are **clearly** off, but only slightly. This will improve, but until then, don't expect perfection. --- ## Comments - **[Radhakrishna MV](https://mvark.blogspot.com/)** _13 Apr 2025 11:50 pm_: I was very impressed that Grok can edit a scanned image of a hand-drawn cartoon & generate a new image - https://mvark.blogspot.com/2025/03/the-diet-delinquent.html This makes it easier to make a cartoon character perform different gestures & movements without having to redraw.