The Nasi Lemak Method to Understanding Generative AI

Heads up: If you buy something through the referral links below, I may earn a small commission (absolutely no extra cost to you). Thanks for supporting my work!

This is the thinking framework I developed for a class on content creation with generative AI.

You’re scrolling TikTok as usual, and you see a cool AI video. And you wonder how to make something similar.

But, how do you start? Why can’t ChatGPT just make it for you? What do all the terms mean, like Higgsfield, Runway, Seedance and Magnific?

If you’re totally confused about this, you’re not alone. Here’s how I break it down to my students using my “Reverse Recipe” method.

Imagine This

Let’s say that you ate a very good nasi lemak at a hawker stall, and wonder how to make the same dish at at home. Well, this is the same as seeing a cool AI video, and thinking how to generate something similar by yourself.

First, you wouldn’t start cooking at random. You’d work out what the dish is, what goes into it, which parts you can buy and which you cook yourself, and what order to cook in.

Basically, there are many ways of making the nasi lemak. In some ways, you make it entirely from scratch, using basic ingredients. Or, sometimes, you can buy parts of it pre-made, like the sambal.

I call this way of thinking Reverse Recipe. It’s “reverse” because you start from the finished dish and work backwards to the ingredients. Then you cook forwards, tasting as you go.

Step 1: Write the dish as one sentence

Nothing gets made until the dish is written down. It’s one sentence, and it answers five questions:

• What am I making, and in what format?
• Who is it for?
• Where will it be posted?
• How long or how big is it?
• What should the viewer think or do afterwards?

So, when you are thinking of using AI to generate a UGC video, for example, don’t just write it as: “A video of an AI influencer reviewing my product”. That’s not enough. You also need to include points like “a 15 second vertical clip for TikTok that gets them to Google my product name”.

Steps 2 and 3: List the ingredients, then break them down

Work backwards from the dish and write down everything it needs. For that AI influencer review, a few things you need to generate it:

• The character i.e. what do they look like
• A script i.e. what are they saying or doing
• The product, and the setting or location

Then take each ingredient and ask if it breaks down into smaller ones. A character might split into a look, a personality and a voice. Add those to your list and ask again, but stop after two levels.

My Singaporean avatar prompt generator is this same breakdown for the look alone. You can’t just say “Singaporean woman” as the character, because you will always get the same character. Try to state the age, build, skin, hair, clothes, expression, setting, light and camera.

Step 4: Decide what you already have, what you buy and what you DIY

Go through every ingredient and sub-ingredient, and give each one a single answer.

Option	What it means	Watch out for
Already have it	You own it or you decide it yourself, like your real product photo, your real price or your audience.	Don’t let AI invent what you already know.
Buy it ready-made	A tool, template or existing asset gives you the ingredient.	You save effort, but you take what it gives you. Check that you can change one part without redoing everything, and that you’re allowed to use the result.
DIY	You make it yourself with a prompt in an AI tool, using a specific AI model and parameters.	It takes more tries, but you control the result and can fix one part at a time.

Remember that whatever you don’t define, the model fills with its most likely guess. That’s how you end up with the wrong price on your ad.

You can mix all three in one dish, like buying the sambal ready-made, cooking the rice yourself and using the egg you already have at home.

A tool that promises the whole dish in one click is a ready-made dish. These could be tools like OpenArt’s UGC creator, or Creatify’s UGC ad creator. It handles everything in one-click. But if you don’t like the result, you usually have to run it again (and again), using up all your credits.

So, pick ready-made when speed matters more than control and the result is easy to redo. Pick DIY when the voice, the accuracy or the legal risk matters.

Steps 5 to 8: Cook in order and taste as you go

Step 5: Plan the steps. Note which ingredients depend on which. One ingredient can depend on several others, and one can feed several others, so draw the links as a web, not a tree.

Start with the sub-ingredients that depend on nothing, and follow the arrows from there. Write one prompt for each DIY ingredient. For a video, make only one shot at first, because the rest repeat the same process.

For example, you must settle on the character’s ethnicity before you generate it. And you must have the generated character before you create the UGC video.

Step 6: Taste. After each output, stop and ask if it gives the next step what it needs. A weak character leads to a weak script and a weak video, so problems grow if you let them through.

If it’s not good enough, fix the prompt and try again, with a fixed limit on attempts so you don’t loop forever.

This is like tasting the sambal before you continue with the rest of the dish. If it’s wrong, it will ruin the final result.

Step 7: Assemble. Join the pieces into one finished dish.

Step 8: QC. Two last checks:

• Matches: does it fit the dish you wrote in step 1, including the audience, the platform, the length and the purpose?
• Allowed: are you allowed to use and share it? Check the rights, make sure you haven’t used a real person’s likeness or a brand’s logo without permission, and label it as AI-made wherever you share it. TikTok’s policy, for one, requires a label on AI-generated content with realistic images, audio or video.

My 22 second home repair video ad has the same shape. I couldn’t one-shot it, so I split it into keyframes, made the images in Nano Banana 2, turned them into clips in Kling, and joined them in CapCut. Some clips took two or three re-rolls before they passed.

Worked example: one Instagram image for a new coffee drink

The framework isn’t only for video. It can apply to pretty much any visuals, such as an Instagram image. Breaking it down:

The dish: One Instagram image for a student cafe, advertising a new iced coffee drink so that students want to try it this week.

Read it from left to right to get the cooking order. The audience and flavour idea are things you already have or decide, so they come first. The name and description are DIY, and the cafe logo is already in the pantry.

The drink image is ready-made from a tool that makes drink posters, and the caption is DIY. Those two come last because both need the finished description.

So you’d cook the name first, taste it, and let it feed the description.

Another worked example: a 30 second microdrama

The dish: One 30 second vertical microdrama for students on TikTok, about a student who catches her roommate eating her leftovers, ending on a cliffhanger so that viewers watch part 2.

The audience and the story idea come first. The script and the character look are DIY. The video clips, voice lines and captions are ready-made from tools.

The script feeds both the voice and the clips, and the character look feeds the clips, so cook those two first.

The taste check that matters here is the script. Read it aloud and time it at 30 seconds before you generate a single clip. Fixing a script costs nothing, but every clip you re-roll costs credits.

For QC, check that no real person is imitated and that the video is labelled as AI-made.

What the Nasi Lemak Method won’t tell you

It gives you a list and an order. It doesn’t tell you which tool to buy, or whether the idea is any good. That’s still your own taste at step 6.

So, you might prefer the avatar created by one tool, such as Gemini. But you might like the video produced by Kling instead of Seedance.

It also doesn’t fix the tools. Even with Nano Banana 2’s consistency features, my character’s face still shifted slightly between frames on the home repair ad. Tasting finds that problem, but it may not make it go away.

So, next time you see something cool and want to create it yourself, write the dish as one sentence before you open any tool.

What’s the last AI video or image you saw and wanted to make? What would be on its ingredient list?