A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK
New Feature

ChatGPT Images 2
Changes Everything

OpenAI just dropped the best AI image generator ever made. Perfect text, full marketing campaigns, readable infographics, and consistent characters across comics. Here are 6 things it can do — with the prompts to try each one.

On April 21, 2026, OpenAI launched ChatGPT Images 2 — a completely new image generation model that fixes the biggest problems AI images have always had. Text that’s actually readable. Layouts that look professionally designed. Characters that stay consistent across multiple images. It’s available right now inside ChatGPT.

Where

Regular ChatGPT chat. Just ask for an image. No special setup.

Plans

Free (3 images/day, basic). Plus ($20/mo) for Thinking Mode + ~200/day.

Speed

60–180 seconds per image. Slower than DALL-E, but dramatically better.

Replaces

DALL-E 2 and DALL-E 3 retire May 12, 2026. This is the replacement.

The Guide 6 Things It Can Do + Prompts to Steal

1. Color Palette From Any Photo

Upload any photo and ChatGPT extracts a complete color palette with hex codes. Use it to build your brand, match a mood, or start a design project with colors that already work together.

Prompt — Color Palette
[Upload your photo] Generate a professional color palette from this image. Create a clean, minimal swatch card showing: - 6 colors extracted from the image (2 dominant, 2 secondary, 2 accent) - Each color as a large swatch circle - The hex code printed below each swatch in clean, readable text - A short name for each color (e.g., "Warm Sand," "Deep Ocean") - White background, modern layout, ready to use as a brand reference

2. Full Marketing Campaigns

Product photography, typography, and layout — all in one shot. This used to require a photographer, a designer, and a copywriter. Now it’s one prompt.

Prompt — Marketing Campaign
Create a premium product advertisement for [YOUR PRODUCT — e.g., "a minimalist ceramic coffee mug"]. The image should look like a professional magazine ad: - Hero product shot with dramatic lighting and soft shadows - Clean background in [YOUR COLOR — e.g., "warm cream" or "deep navy"] - EXACT headline text: "[YOUR HEADLINE — e.g., "Made to Last. Made for You."]" - Subheadline text: "[YOUR SUBHEADLINE — e.g., "Handcrafted in Portland. $38."]" - Small logo or brand name in the bottom corner: "[YOUR BRAND NAME]" - Modern sans-serif typography, high contrast, editorial feel - Aspect ratio: 4:5 (Instagram portrait) This should look like it belongs in a high-end lifestyle magazine. Not AI-generated looking — polished, intentional, real.

3. Magazine Layouts With Readable Text

This is the big one. Text in AI images has always been garbage — misspelled, warped, unreadable. Images 2 renders long passages with perfect spelling and consistent formatting. You can create actual magazine pages, posters, and menus.

Prompt — Magazine Layout
Create a modern magazine feature page layout about [YOUR TOPIC — e.g., "the rise of solo travel"]. Include these EXACT text elements — spell every word perfectly: - Magazine title at the top: "[e.g., WANDERLUST MAGAZINE]" - Article headline: "[e.g., The New Solo Traveler: Why More People Are Going Alone]" - Byline: "[e.g., Words by Sarah Chen | Photography by James Liu]" - Pull quote in large text: "[e.g., I stopped waiting for someone to go with me. That's when the real trip started.]" - 2-3 paragraphs of body text in a realistic two-column layout (generate realistic article text about the topic) Style: Clean editorial design. Mix of photography and text. Professional typography with clear hierarchy — title largest, pull quote emphasized, body text small but readable. Think Kinfolk or Cereal magazine aesthetic. Full bleed page, no white border.

4. Realistic Handwriting

AI used to be terrible at handwriting. Now you can create handwritten notes, journal entries, and letters that look genuinely real. Great for social media content, product mockups, or just fun.

Prompt — Handwritten Note
Create a photograph of a handwritten note on lined notebook paper, slightly angled on a wooden desk with a pen beside it. Natural lighting, casual feel. The note should be written in neat but natural handwriting (not perfect — slightly imperfect like real human writing). The EXACT text should read: "[YOUR TEXT — e.g., Things I'm grateful for today: 1. Morning coffee on the balcony 2. That conversation with Mom 3. Finally finishing the book 4. The way the light hit the kitchen at 4pm Tomorrow's going to be a good day.]" The handwriting should look like a real person wrote this in their journal. Blue ink. Slightly different letter sizes. Natural spacing. Not a font — actual handwriting style.

5. Full Infographics

Research breakdowns, charts, stat cards, comparisons — all designed together in one image with readable text. This used to require Canva or a designer. Now it’s a prompt.

Prompt — Infographic
Create a clean, modern infographic about [YOUR TOPIC — e.g., "The State of Remote Work in 2026"]. Include these EXACT stats and text — spell every word perfectly: Title: "[e.g., Remote Work in 2026: By the Numbers]" Stats to include: - "[e.g., 42% of US workers now work remotely at least 3 days/week]" - "[e.g., Average salary premium for in-office roles: $8,200]" - "[e.g., Top remote industries: Tech (68%), Finance (52%), Marketing (47%)]" - "[e.g., 73% of remote workers report higher job satisfaction]" - "[e.g., Companies offering remote: 2019 = 16% → 2026 = 58%]" Design requirements: - Vertical format (1080x1920 — Instagram Story / Pinterest) - Clean color palette: [e.g., "navy, white, and coral accents"] - Mix of icons, simple bar charts, and large stat callouts - All text must be perfectly spelled and clearly readable - Source line at the bottom: "[e.g., Source: Bureau of Labor Statistics, FlexJobs 2026]" - Professional, shareable, ready to post

6. Comics With Consistent Characters

The biggest pain in AI art has been character consistency — the same person looking completely different in every image. Images 2 with Thinking Mode can generate up to 8 coherent images in one batch, keeping characters locked.

Prompt — Comic Strip
Create a 4-panel comic strip in a clean, modern illustration style. Character: [Describe your character — e.g., "A woman in her 30s with short dark hair, round glasses, wearing a blue oversized hoodie. Warm skin tone, friendly expression."] This character must look EXACTLY THE SAME in every panel — same face, same hair, same outfit, same glasses. Character consistency is critical. Panel 1: [e.g., She's sitting at her desk staring at a laptop, looking frustrated. Speech bubble: "Why won't this code work?"] Panel 2: [e.g., She opens ChatGPT on her phone. Speech bubble: "Let me ask ChatGPT."] Panel 3: [e.g., Close-up of her face looking amazed at her phone screen. Speech bubble: "Wait... that actually worked?"] Panel 4: [e.g., She's leaning back in her chair with a huge smile, arms behind her head. Speech bubble: "I'm basically a developer now."] Style: Clean lines, bright colors, slightly exaggerated expressions (not anime — more like a modern webcomic). White background between panels. Speech bubbles with readable text. Arrange as a 2x2 grid.
Tips Get Better Results

Use “Exact Text” Formatting

When you want specific words in the image, write them in quotes and say “EXACT text” or “spell every word perfectly.” This tells the model not to paraphrase or improvise. For tricky words, spell them letter by letter.

First 50 Words Matter Most

The model weights the beginning of your prompt more heavily. Lead with the most important details: what type of image, what’s the subject, what style. Save fine details for the end.

Say “No Extra Text”

The model sometimes adds words you didn’t ask for. If you only want specific text in the image, add: “Do not add any text besides what I specified.”

Thinking Mode = Better Everything

If you’re on Plus ($20/mo) or higher, the model uses Thinking Mode automatically for complex requests. It plans the composition before generating. This is why it’s slower (60–180 seconds) but dramatically better. Free users don’t get this.

Iterate in the Same Chat

Unlike DALL-E, Images 2 remembers your conversation. Say “make the background darker” or “change the headline to...” and it edits without starting over. Stay in the same chat for the best results.

Honest Limitations

Slower than DALL-E — 60–180 seconds vs. 20–45 seconds. Worth the wait, but not instant.

No transparent backgrounds — Unlike the previous model, Images 2 can’t generate transparent PNGs.

Free tier is limited — ~3 images/day without Thinking Mode. Plus ($20/mo) gets ~200/day with Thinking Mode.

Still struggles with complex physics — Origami folds, Rubik’s Cubes, and intricate mechanical parts can be hit or miss.

Editing has limits — A bad first image rarely recovers through prompting alone. If the first result is off, start a new generation instead of iterating endlessly.

Images 2 is incredible for visual work. But the real productivity upgrade comes from building a full AI system around your job — automating your most repetitive tasks, connecting AI to your email and calendar, and building skills that fire with one sentence.

The Only AI Masterclass You Need

Build AI Systems That Run Your Work, Business, And Life

If this guide helped, but you’re looking to go deeper, I got you!! My 30-Day Challenge takes you from saving AI tips you never use to actually building with AI, step-by-step.

I show you exactly how I automated two e-commerce brands, my social media, and most of my personal life, then hand you the agents, workflows & systems to do the same. I’m teaching you every single thing I know with one lesson and one build a day.

Join the AI Masterclass →

© 2026 Mariah Brunner. All rights reserved.