It was a pool of insights in this conversation between @AlexHormozi and @harleyf on alex hormozi's podcast.
Here are some insights I found essential to know for everyone.
What's your favorite lesson from this podcast? Let me know below! 👇
While you argue about which model is "the best," the people out-building you are using all of them.
Claude Code. Codex. Antigravity.
Not because they're indecisive.
Because creativity shouldn't be capped at one tool's ceiling.
This combination is a straight away a winner:-
References + Stitch + Antigravity + Claude Code + Codex
If you know the correct way to juggle between all these tools to get your desired output, you're already in the top 0.1% users of AI.
Stop restricting yourself to a single model.
Instead, try to win by using them all.
Most people don’t get bad AI thumbnails because the model is weak.
They get bad thumbnails because their input is weak.
People say:
“GPT Images 2.0 / Nano Banana 2 can’t create unique thumbnails.”
But I’ve seen people generate insane visuals with the same models.
The difference?
The prompt
Your prompt should define:
→ emotion
→ composition
→ lighting
→ text placement
→ visual style
→ subject clarity
The reference
This is the real cheat code.
A good reference tells the model what “premium” looks like.
Most people only prompt the AI.
The best creators direct it.
Prompt = what to create.
Reference = what good looks like.
That’s why two people can use the same model and get completely different results.
What do you struggle with most in AI thumbnails?
Realism, layout, text, or uniqueness?
@siddharthwv Your results depend on the kind of knowledge you hold and the value you share.
Strong knowledge and value always create wealthy offers.
Nice work bro!!
This combination is a straight away a winner:-
References + Stitch + Antigravity + Claude Code + Codex
If you know the correct way to juggle between all these tools to get your desired output, you're already in the top 0.1% users of AI.
Stop restricting yourself to a single model.
Instead, try to win by using them all.
Most people don’t get bad AI thumbnails because the model is weak.
They get bad thumbnails because their input is weak.
People say:
“GPT Images 2.0 / Nano Banana 2 can’t create unique thumbnails.”
But I’ve seen people generate insane visuals with the same models.
The difference?
The prompt
Your prompt should define:
→ emotion
→ composition
→ lighting
→ text placement
→ visual style
→ subject clarity
The reference
This is the real cheat code.
A good reference tells the model what “premium” looks like.
Most people only prompt the AI.
The best creators direct it.
Prompt = what to create.
Reference = what good looks like.
That’s why two people can use the same model and get completely different results.
What do you struggle with most in AI thumbnails?
Realism, layout, text, or uniqueness?
There are ideas that you think will work.
And there are some ideas that will actually work.
Once you get the difference
Your implementation will make you more confident.
Paste the report into the next model.
It picks up exactly where the last one left off.
Net result: zero downtime, even on free tiers.
Limits are a workflow problem, not a credit problem.
When one runs out of credits, don't wait.
Ask it to write a "Handoff Report" before you close it:
– What we built
– What's still broken
– What the next model needs to know
A LinkedIn profile audit in 4 questions:
1. Does your headline name your buyer, or your job?
2. Does your About open with a story, or a list of services?
3. Is your featured section empty, or showing real proof?
4. Does your banner sell, or just sit there?
Most founders fail 3 of 4.
@KarolAnurag Just to inform you that I'm the wrong Anant Goel you're complaining to. I guess if you complain to the right person, you'll get your problem solved.
Step 4
Copy the generated prompt from Claude.
Open ChatGPT Images 2.0
Provide 3 - 5 photos of yourself where you're clearly visible.
Paste the prompt, wait for a few seconds.
And here it is, you successfully generated your first photo of your business Photoshoot.
Repeat for different poses.
Test it out and see what works.
Before you start learning to design, you have to learn how references are used in any creative industry.
Knowing how to use references is a superpower for someone working in the creative industry.
Step 4
Copy the generated prompt from Claude.
Open ChatGPT Images 2.0
Provide 3 - 5 photos of yourself where you're clearly visible.
Paste the prompt, wait for a few seconds.
And here it is, you successfully generated your first photo of your business Photoshoot.
Repeat for different poses.
Test it out and see what works.
Most people still think you need a photographer for a business photoshoot.
You don’t.
With ChatGPT Images 2.0, you can generate a full studio shoot that looks exactly like you.
↓↓↓ Here’s how to do it ↓↓↓