Most people's first AI image generation experience goes like this.
They open Midjourney or ChatGPT. They type something like 'a beautiful product photo of my skincare serum'. They get back something that looks vaguely like a serum bottle but nothing like what they imagined. They try again with slightly different words. Still not right. They close the tab and conclude that AI image generation is overhyped.
The problem is not the tool. The problem is the approach.
AI image generation rewards a specific way of thinking and describing. Learn it once and you get good results almost every time. This guide teaches you that approach and gets you to a good result in under 10 minutes.
The mindset shift that changes everything
Stop thinking of AI image generation as a search engine where you type what you want and the best result appears.
Start thinking of it as giving instructions to a photographer who has never met you, never seen your product, and has no context for what you are trying to create. That photographer needs to know everything - not just the subject but the lighting, the composition, the mood, the camera, the surface, the colors.
The more specific your instructions the more accurately the photographer executes your vision. Vague instructions produce generic results. Specific instructions produce specific results.
This is the entire secret of AI image generation. Specificity is the skill.
The 3-layer minimum prompt
You do not need to master the full 6-layer prompt framework on your first attempt. You need 3 layers minimum to get a good result.
Layer 1 - What it is: define the type of image before anything else. Not just 'a photo' but a specific type. Product still life. Portrait. Fashion editorial. Flat lay. Interior scene. This single decision frames everything that follows.
Layer 2 - What is in it: describe your subject with at least 3 specific details. Not 'a serum bottle' but 'a dark amber glass serum bottle with a gold dropper cap, sitting on a white marble surface with warm gray veining'. Three details minimum - material, color, and surface or context.
Layer 3 - How it is lit: add one lighting sentence. This is the most impactful addition you can make to any prompt. 'Soft diffused natural window light from the left, warm golden undertones' transforms a flat generic image into something that feels real and intentional.
That is your minimum viable prompt. Three layers. One sentence each. Already dramatically better than most first attempts.
Your first prompt - copy and customize this
Here is a template you can use right now. Replace the bracketed sections with your specific content.
'Product still life of [describe your product - material, color, shape, any distinctive features], sitting on [describe the surface - white marble, warm wood, linen cloth, concrete], [describe any props or context if relevant - otherwise leave blank], soft diffused natural window light from the left with warm golden undertones, clean minimal background, shot on 85mm lens at f/2.0'
Real example using this template:
'Product still life of a dark amber glass serum bottle with a matte gold dropper cap, sitting on a white marble surface with warm gray veining, soft diffused natural window light from the left with warm golden undertones, clean minimal background, shot on 85mm lens at f/2.0'
If you are using Midjourney add these parameters at the end:
'--ar 4:5 --v 8.1 --stylize 150'
If you are using ChatGPT just paste the description text with no parameters.
Run this prompt now. Your result will be significantly better than your first attempt because it has all 3 minimum layers.
What to do if the result is not quite right
Your first generation will probably be close but not perfect. That is normal and expected. Here is how to fix the most common problems quickly.
The colors look wrong: add more specific color language. Instead of 'amber' try 'deep warm amber with golden undertones, translucent when light passes through'. Instead of 'gold' try 'warm brushed gold with a matte satin finish'.
The lighting looks flat: make your lighting description more specific. Add the position explicitly - 'from the upper left at 45 degrees'. Add the shadow quality - 'gentle graduated shadow falling to the right'. Add the mood - 'warm and intimate quality'.
The background is cluttered or wrong: add explicit background instructions. 'Seamless white background with subtle center gradient, no visible edges or props' or 'soft blurred warm interior background, no specific details visible'.
The product looks wrong or generic: add more product details. Describe the exact shape, the finish of the material, any distinctive design features. The more specific you are about the product the more accurately the AI renders it.
The image looks too AI-generated: add '--raw' at the end of your Midjourney prompt. And add camera-specific language - 'shot on Canon EOS R5, natural skin texture, visible material grain, slight natural imperfections'.
The composition is always centered: add explicit composition direction. 'Product positioned on the left third of the frame, generous negative space to the right, rule of thirds composition'.
The 10-minute plan
Minute 1: write your first 3-layer prompt using the template above. Do not overthink it. Get something written and run it.
Minutes 2 to 3: run the prompt. Review the 4 results. Identify the best one and what specifically is not right about it.
Minutes 4 to 5: make one specific improvement to the prompt based on what you observed. If the lighting was flat make the lighting description more specific. If the colors were wrong make the color description more specific. Change one thing at a time.
Minutes 6 to 7: run the improved prompt. Review the results. Select the best image.
Minutes 8 to 9: upscale your selected image to full resolution. Download it.
Minute 10: evaluate honestly. Is this image good enough to use? If yes you are done. If not identify one more specific improvement and run one more generation.
Most people get a genuinely usable image within 2 to 3 generations using this approach. The key is making one specific targeted improvement per iteration rather than rewriting the entire prompt.
The single most common beginner mistake
Rewriting the entire prompt every time a generation does not work.
When you change everything at once you cannot tell what made the difference. You lose the parts that were working and have no information about what specifically caused the problem.
The professional approach is surgical iteration. Identify the specific element that is wrong. Change only that element. Run again. Observe the difference. This approach produces consistent improvement and builds your understanding of how prompt language affects output.
One change per iteration. Every time.
What to learn next
Once you have your first good result you have proven the fundamental concept. The next step is understanding the other 3 layers of the full 6-layer framework that you did not use in your first prompt.
Layer 4 - Composition: how the image is framed and where things are placed in the frame.
Layer 5 - Colors and materials: specific descriptive language for colors and surface materials.
Layer 6 - Camera and lens: focal length, aperture, camera look.
Each additional layer gives you more control and produces more professional results. You do not need all 6 layers for every image but understanding all 6 gives you the ability to fix any specific problem that appears in your output.
The fastest way to apply all 6 layers without memorizing the framework is to use a tool that builds them for you.
Our free AI Image Prompt Generator lets you describe what you want in plain English and generates 5 ready-to-use prompt variations - safe, cinematic, editorial, minimal, and experimental - each built using the full 6-layer framework. Copy any of them directly into Midjourney or ChatGPT.
If you want to build prompts layer by layer with live preview the Midjourney Prompt Builder lets you select each element from dropdowns and watch the prompt assemble in real time.
Open the Prompt Builder →
CropUp Ai