TL;DR: ChatGPT Images 2.0 dropped on April 21. Text that actually renders. Thinking mode that actually researches. I cancelled my subscription a month ago, and now I’m sitting here with six browser tabs open reconsidering that decision. Here’s what I tested, what actually impressed me, and the prompt pack I built for Substack creators.
A month ago I cancelled ChatGPT. I was paying for it, not using it, and Claude had become my main workspace. Then OpenAI dropped Images 2.0 on April 21 and I watched my timeline lose it.
I’d seen the hype before. I was not going to get burned again by an AI image tool that turns every piece of text into decorative gibberish.
So I tested it myself. Specifically for Substack creators. Not for fun, not for art — for the actual stuff we need week to week.
Here’s what happened.
I wrote about Claude Design when Anthropic released their visual tool. Check out that post here if you want the full comparison context.
Hey, I’m Pinkie. On this corner of the internet, we build your entire visual brand system using AI, one step at a time. If you’re new here: welcome. Honestly, this might be the best time to join. start here! 🩷🦩
What you’ll learn in this post:
Why ChatGPT Images 2.0 is different from every image generator you’ve tried before (and why text rendering is the headline)
How to use it as a Substack creator specifically — covers, infographics, social posts, and more
Three prompts you can steal right now
The full 7-prompt Substack creator pack, formatted and downloadable (paid)
What Changed in ChatGPT Images 2.0?
Before we get into prompts, here’s why this update matters.
Text rendering is the headline. Previous image AI treated text like a visual texture — it looked like letters if you squinted. Images 2.0 renders clean, readable, accurate text in small UI labels, captions, headlines, and non-Latin scripts. This is not a small improvement. It makes infographics actually possible.
Thinking mode lets ChatGPT search the web before generating. Give it a Substack URL and it reads the article, understands the theme, then generates. This is the capability that made my jaw drop.
Precision is harder to describe but easier to feel — when you give it specific layout, style, and content instructions, it follows them. The fine-grained control previous models fumbled is noticeably better.
Important note before you dive in: ChatGPT Images 2.0 lives in Thinking mode and is only available for paid ChatGPT users. Instant mode is free. For the prompts below that use article URLs, you need Thinking mode.
1. The Infographic Prompt
I’ve watched creators like Karen Spinner (I tried using Nano Banana Pro to create complex infographics) and Daria Cupareanu (58 Nano Banana 2 Prompts for Infographics) both trying to crack the infographic problem a few months ago and nobody really found a good solution. And then ChatGPT comes along and… I am impressed. The text is perfect, not a typo in sight. I used a very simple prompt to generate an infographic about my Claude Design setup tutorial post and it just... worked.
This is where I stopped and stared at my screen. I was so surprised by the quality, I half convinced myself it was so good because ChatGPT had learned my brand by now. So I tried it on someone else's post and branding that my AI doesn't know yet. I gave it Karen Spinner and Dr Sam Illingworth’s post Does it matter if you’re polite to Claude?
Here’s what I got:
Verdict: I’m not overstating this, infographics are now a viable Substack asset. That wasn’t true two months ago. I got a clean infographic with readable text, no typos, sensible layout. Apart from Karen’s double hand, visual control is also stronger than before.
Prompt:
Read this post and create an infographic that helps the reader understand the main message: [URL]
Create the infographic imitating the style of the attached reference images. [ATTACH STYLE IMAGES] 
If today’s post made you think about your own visual brand, I am launching The Branding Flock, a 12-month visual brand program for Substack creators who are done guessing.
Full breakdown posts, real prompts, real outputs. Your color palette, your logo, your brand character, your content system — built step by step alongside the flock.
33% off closes after the first 100 seats. The discount lives inside your quiz result.
2. Social Media Post Batch
Another category that genuinely surprised me. I work at a creative agency — getting our designers to create eight social media posts takes a week. ChatGPT created eight in five minutes. And six out of eight I’m actually thinking of posting.
It took my Substack page and, using my articles, created posts for each of them. On-brand colors, captions that make sense, the kind of thing I’d consider running as paid promo on Instagram and Facebook.
And here‘s what I got testing it on Mia Kiraki 🎭’s brand ROBOTS ATE MY HOMEWORK:
Verdict: Girlboss approved.
Prompt:
I am a Substack creator who shares content about [TOPIC] with my audience. My website/Substack is: [INSERT LINK]. Review my website and create 8 different social media post designs for Facebook and Instagram.
Make the posts distinct from each other. Base them on my existing content, themes, tone, visual design, and audience.Tag me if you try any of these — I'm AI Meets Girlboss everywhere. I want to see your outputs, and I'm restacking every experiment this community shares. If this is saving you time, send it to one person who needs it. Thank me later. 🩷
3. Forbes Double-Feature Editorial
It’s actually crazy. Give ChatGPT a link to your website and a photo, and it does the research with a very simple prompt — and we’re used to seeing AI-generated text as gibberish, but here it’s not just without typos, the content actually makes sense.
ChatGPT really got their game together. Take a look at the Forbes article it generated purely based on my Substack link and a portrait photo. A full double-page spread. Headline, Q&A, pull quotes, editorial layout. Consistent likeness. Readable copy.
Here’s what it did for Kim Doyal, just by giving ChatGPT Images 2.0 the prompt below:
Prompt:
Research [NAME/BRAND] on [PLATFORM] and, using the attached image as the identity reference, create a double-page premium business magazine feature.
Include: headline + subheadline, short researched feature article, 4-question interview section, editorial photoshoot images based on the attached photo, pull quotes, captions, and a polished layout. Make it feel like a real high-end editorial spread, not an ad, with consistent likeness and readable copy.4. Logo Generator Prompt
OK so I’m picky. Ask anyone who’s worked with me on branding. I have opinions.
First attempt: I gave it my Substack URL and asked for five logo variations. The output was fine. Not groundbreaking, a bit dull, the kind of thing you’d accept if you had no other option. I was not impressed.
But here’s what I did instead of giving up — I added more constraints. Specific colors, symbols, style direction. Second time around I got something actually worth looking at.
I even tested it on my new Front Row client, Badass Women in AI, and gave it a bit more specific brief about colors, symbols and style.
This is the thing about Images 2.0 that I keep coming back to: it rewards specificity. Vague brief, vague output. The more you give it, the more it gives back. Which, honestly, is just good briefing. Same rules apply whether you’re working with AI or a human designer.
Not replacing your brand designer. But as a starting point for visual direction? Useful.
The prompt for logo generation in ChatGPT Images 2.0 is in the downloadable pack at the end of this post.
Want someone to build your visual brand with you?
The Front Row is my 1:1 brand-building program — 30 days, your visual brand, done. We define your foundations, build your visual world, and set up a system that keeps you consistent long after we’re done.
If that sounds like what you need, DM me or apply below — I take a limited number of clients at a time.
5. The Substack Cover Image Prompt
This was the most practical test. I regenerated a cover for one of my existing posts about creating video-style covers for Substack posts.
Tried it on Tracy Friedlander’s 3 Substack Notes Trends post too.
Verdict: not perfect. Neither my fashion illustration nor Tracy’s comic book style fully transferred. The colors and linework don’t match my AI Meets Girlboss visual system the way Gemini does. True, I’m very picky. I do appreciate the text rendering though, which was done entirely by ChatGPT, without me prompting anything text-related.
But(!) if you had two minutes and needed something that wasn’t stock imagery? It works. It reflects the post. It doesn’t look like a Canva template.
If you want deep visual brand storytelling, that’s a different workflow. If you need a cover that isn’t embarrassing in the next 10 minutes, this does it.
The prompt for Substack cover generation in ChatGPT Images 2.0 is in the downloadable pack at the end of this post.
6. Comic Book-style Post Summary
This was fun. I’ll be honest, I’m a visual person, and reading long articles makes me tired sometimes. I know, not the popular thing to say on Substack. But that’s why I appreciate writers like Mia Kiraki 🎭 who add visuals that highlight the important messages (check out this post, it’s a visual reader’s dream!).
I thought it would be fun to create a TLDR visually, and since OpenAI lists storyboards for creatives as a feature, I recreated my 0 to 450 subscribers Q1 performance review post as a one-page comic book.
And Manisha wrote a piece on how she uses Duck.ai as a safer sandbox for family AI experiments, which just seemed like the perfect comic book experiment.
It really delivered for both brands. Both visually and content-wise — the character held across panels, the story tracked, the text is clean. Genuinely useful for visual learners and for repurposing a post for visual platforms, like Instagram.
The prompt for comic book generation in ChatGPT Images 2.0 is in the downloadable pack at the end of this post.
7. Screenshot Mockup
I use a lot of screenshots in my posts and I used to put them in Canva or Photoshop (I know, old school) to add a frame, so it looks on-brand. I didn’t trust AI with them because it would just take the text on the screenshot and hallucinate something.
I decided to challenge it by giving it not one, but five screenshots.
Now look at this — a hot pink frame and ALL five screenshots are intact. Took me 30 seconds versus 10 minutes.
The prompt for screenshot generation in ChatGPT Images 2.0 is in the downloadable pack at the end of this post.
My Verdict on ChatGPT Images 2.0
ChatGPT Images 2.0 is good enough for solopreneurs who have a visual style, color palette, and vibe they want to maintain.
If you’re a solo creator building weekly content and you have a general aesthetic in mind, ChatGPT 2.0 can deliver that consistently. You’ll save 3–4 hours a week on image generation. That’s real ROI.
But here’s where it stops: If you’re a professional brand with exact rules—specific kerning on logotypes, pixel-perfect paper grain, locked color values, rigid visual systems—this isn’t your tool. Nano Banana is. Gemini gets closer. ChatGPT 2.0 still isn’t there.
Text rendering: 9/10. Infographics are now real. This alone justifies the update — and the subscription.
Thinking mode: 8/10. The URL research is something else. Needs a strong style reference to land — don’t skip it.
Precision and control: 7/10. Better than anything before it. Still not pixel-perfect for tight brand systems. Round two helps.
Did I resubscribe? Still deciding. But I have six new tabs open, which says something. And knowing how fast this space moves — Gemini will probably drop something new in two days anyway.
In the comments…
Which of these would you try right away? Cover images, infographics, or the social media batch? Drop it below, I’m curious (as always).
Want all seven prompts from this post, ready to copy?
The downloadable prompt pack has everything — prompts, mode notes, and usage tips — without hunting through this post. Upgrade to get the download!
