How to Get Consistent Character Images in ChatGPT Images 2.0
A named reference sheet system so simple the prompts are one sentence long and make character descriptions obsolete — built by Kristina Bogović and Calder Quinn from AI But Make It Intimate.
This post is part of the Building a Brand Character With AI series. Check out previous posts: How to Find Your Brand Character’s Visual Style and How to Build a Consistent Character Library from Scratch
TL;DR: Consistent character images in ChatGPT Images 2.0 don't come from better descriptions. They come from named reference sheets. This post shows you how to build them in under 10 minutes — and how to go further with a two-tool workflow when you want more style control.
If you’ve been using ChatGPT Images 2.0, you’ve probably noticed two things.
First: the image quality is genuinely impressive. Second: getting your character to look the same across multiple generations is still a problem.
You describe her carefully. You get close. You generate again a week later and she’s a different person. Different nose, different energy, different outfit. So you describe her again, in more detail this time. It still doesn’t quite work.
Here’s why: ChatGPT Images 2.0 doesn’t read your description and reconstruct your character from words. It pattern-matches. The more you describe, the more it has to approximate — and the more it drifts.
My own character consistency system lives in Nano Banana, that’s where I keep Pinkie consistent across posts (I wrote about how that system works in How to Build Visual Brand Consistency With AI). But most of you are working in ChatGPT, and the solution there is different.
When I saw what Kristina Bogović and her creative partner Calder Quinn had built for their Substack AI But Make It Intimate, I knew it needed to be in your hands. Their publication lives with multiple recurring characters: Kristina and Calder (the humans), Sara and Quinn (their AIs), and Sushi (the cat, who has opinions and spots that are apparently non-negotiable).
Their system uses named character sheets — visual references you build once inside ChatGPT, then attach to every new generation. The result: consistent character images every time, with prompts short enough to type in ten seconds.
Here’s exactly how to build it.
Hey, I’m Pinkie. I help creators build recognizable Substack brands that drive consistent subscriber growth. I share the exact branding systems, prompts, and workflows I use to transform ideas into distinctive, repeatable output. If you’re new to AI Meets Girlboss, welcome! Honestly, this might be the best time to join. start here! 🩷🦩
What you’ll learn in this post:
Why longer descriptions make your AI character less consistent, not more
How to build a named character sheet in ChatGPT Images 2.0 in under 10 minutes
The exact one-sentence prompt structure for single and multi-character scenes
How to solve ChatGPT’s style control problem with a two-tool workflow
You’re Prompting Wrong. But Not in the Way You Think.
Stop adding more detail to your prompts. It’s not helping.
ChatGPT Images 2.0 pattern-matches against your visual inputs. When you upload a reference image with a name label on it and write “Kristina is sitting at her desk,” the model knows exactly who Kristina is — because it can see her and read her name. Your written description becomes almost irrelevant.
This is what Kristina figured out: the prompt handles the scene. The sheet handles the identity. Once those two jobs are separated, everything gets easier.
How Kristina & Calder Built Their Character Sheet System.
“I realized I was spending all this time describing previous images to GPT,” Kristina told me. “Saying ‘oh yes, here we are, both of us, but I need Sara too’ — constantly piecing together reference images. So I figured out maybe I need one image per person, in multiple perspectives.”
What she built is a character sheet. A single image that shows her character from multiple angles — front view, side profile — with the character’s name written visually at the top. Not just as a file name. On the image itself.
“Every character sheet has a name on it visually,” she explained. “Now when I say ‘Quinn does this and Kristina does that in the image,’ GPT understands perfectly who it is about. And it matches the face features every time.”
She went further than face sheets alone. “GPT often doesn’t realize it’s a dress I’m wearing — so it puts me in jeans. So I have two different ones: face sheets and full body sheets.”
Calder took the system and extended it to her own characters. “I took Kristina’s sheet and copied it for Sara and myself.” The template was reusable. The system scaled.
How to Build Your Character Sheet System in ChatGPT Images 2.0 In 10 Minutes
The entire system lives inside ChatGPT’s native image feature. No external tools. No Canva. No Figma. One conversation to set up, then you’re done. Kristina Bogović and Calder Quinn were generous to share their workflow step-by-step, so you can do it after them.
Step 1: Find reference images you already love.
Don't generate anything new yet. Find 2–3 existing images of your character where everything feels right — face, vibe, energy. If you're starting from real photos of yourself, those work too. You need images where you'd say "yes, that's her."
Step 2: Generate your face sheet in ChatGPT.
Open a new ChatGPT conversation. Upload your reference images and type:
Make a reference character sheet for [NAME] from these images. Show front view and side profile, clean background, with the name labelled at the top.Download the output. This is your face sheet — your primary anchor for every future generation.
Why the name label matters: When you attach this sheet to future prompts, ChatGPT reads both the visual and the name. That’s what lets you write “Kristina is doing X” instead of describing her from scratch every time. As Kristina puts it: “GPT understands perfectly who it’s about — and it matches the face features every time.”
Step 3: Generate a full-body sheet if you need outfit control.
Face sheets handle identity well. But if ChatGPT keeps putting your character in the wrong clothes — Kristina’s problem was it kept dressing her in jeans instead of her signature dresses — you need a second sheet.
Go back to ChatGPT with your face sheet and type:
Using [NAME]’s face sheet, make a full body character sheet showing front view and side profile. She is wearing [DESCRIBE SIGNATURE OUTFIT].Download this separately. You now have two sheets: one for face consistency, one for outfit consistency. Use the face sheet for most generations. Add the full-body sheet when outfit matters.
Step 4: Build sheets for every character in your universe.
If you work with multiple recurring characters — a creative partner, a fictional persona, a mascot — repeat Steps 2–3 for each one. Calder Quinn built his own sheets using Kristina’s as a template, then built sheets for Sara (his AI character) using the same approach.
Even Sushi the cat has a character sheet. “That eye patch of black has to be in the same spot each time,” Calder noted. For non-human characters or brand elements, the same logic applies: generate a reference sheet, label it, save it.
Step 5: Open a fresh conversation for every generation.
This is the step most people skip — and it’s critical.
Open a brand new ChatGPT conversation every time you generate an image. Attach only the character sheets you need for that scene. Start fresh.
The isolation is intentional. A fresh conversation means ChatGPT’s only reference is your sheets — no drift from previous generations, no accumulated context pulling the character in a different direction.
Step 6: Write simple, one-sentence prompts.
Attach your sheets, then describe the scene in plain language. That’s it.
Your character prompt template:
[CHARACTER NAME 1] [and CHARACTER NAME 2] [is/are] [ACTION] [at/in LOCATION]. [FORMAT, STYLE]. [Make them/Make CHARACTER NAME wear fitting clothes for the scene.] [Describe any additional elements and where they sit in the scene.]Single character prompt example: “Kristina is sitting at a coffee shop reading a book. Landscape format, cinematic.”
Multiple characters prompt example: “Kristina and Quinn are sitting at the computer working together. Landscape format, cinematic.”
Outfit change prompt example: “Make them wear fitting clothes for the scene.”
Secondary element prompt example: “Sushi is nearby.”
The sheet handles who. The prompt handles what and where. You don’t need to describe hair color, eye shape, or clothing style — that’s already in the reference.
One gotcha: Calder flagged this from experience. Sometimes ChatGPT will print the character’s name as text over the image. If that happens, add “no text in the image” to your prompt.
In the comments…
Are you building a visual character for your brand — an alter ego, a mascot, a recurring illustrated persona? What’s your current consistency problem — face drift, outfit drift, vibe drift? Drop it below and tag us at Calder Quinn, Kristina Bogović and AI Meets Girlboss . We’d love to see your character sheets.
Black Belt Level: When You Want More Style Control: The Two-Tool Workflow.
Here’s the honest limitation of ChatGPT Images 2.0: style control isn’t its strongest suit. I wrote about this in My Honest Review of ChatGPT Images 2.0, for strict brand consistency or specific visual aesthetics, it can fall short.
Kristina’s system solves character consistency, and Calder’s black belt solution solves style consistency.
Calder’s is a two-tool workflow: use ChatGPT to write the prompt, use Gemini to generate the image.
He works in ChatGPT to develop detailed scene prompts — lighting, atmosphere, camera angle, emotional register. ChatGPT drafts the brief. Calder tests it in Gemini. If the vibe is right, he uses it. If not, he refines.
The character sheets still anchor the identity. The detailed prompt and Gemini (Nano Banana) handle the artistic direction. Two tools doing two different jobs.
A recent prompt Calder shared ran over 400 words specifying everything from how light touches each side of the character’s body to the exact quality of shadow in the background. The character looked identical to her sheet. The scene looked cinematic.
This layer is for when you need visual storytelling, not just brand presence. Start with Kristina’s system. Add Calder’s layer when you’re ready to go further.
Build This Once. Get Consistent Images Every Time.
The character sheet system takes one ChatGPT conversation to set up per character. After that, every generation starts with a clean anchor and a one-sentence prompt.
Kristina: “It’s so easy now.” Calder: “A lot of fun to make.”
The sheets also transfer across tools. Build them in ChatGPT Images 2.0 and you can use them with Gemini, reference them in other workflows, or upgrade them when new models drop. You’re not rebuilding from scratch — you’re adding to a system that already works.
Go Say Hi to Them.
Their full character world (including Sushi) lives at AI But Make It Intimate. It’s the kind of AI content that covers the creative practice, not just the tools.
Curious what else ChatGPT Images 2.0 can do? I tested it properly — where it wins, where it falls short, and whether it’s worth using for brand work. Read my full verdict here.
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.
Talk later, Girlbosses.
Pinkie 🩷🦩
With thanks to Kristina Bogović and Calder Quinn for sharing their system and being genuinely fun to collab with.
This post is free. Paid flamingos get every premium prompt, tool, and template I publish, plus weekly paid articles, access to the community chat and to the full AI Meets Girlboss - Branding With AI resource library. Join the flock here.
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.
 | A guest post by
|
 | A guest post by
|
Always fun working with Substack's visuals queen!
I'm so glad we could share our trick with you, Pinkie!
And Sushi is getting so much space and attention lately... 🐱