The single most underrated capability of gpt-image-2 isn't the text rendering everyone's writing about. It's multi-image coherence. With Thinking Mode enabled, you can generate up to 8 visually consistent images from a single prompt — same character, same color grading, same aesthetic, same brand feel.
For social media in 2026, this is the unlock. LinkedIn carousels (10 slides) and Instagram carousels (4-10 slides) have been outperforming single-image posts for over a year. The bottleneck wasn't strategy — it was production. Designing 10 visually consistent slides used to take a half-day for a designer or $500 to a freelancer. With gpt-image-2 it's one prompt and a Figma composite session.
This article shows you exactly how. It assumes you've read the foundational gpt-image-2 review — particularly the documented weaknesses, because carousel-generation amplifies some of them in interesting ways.
---
Why Carousels Outperform Single Posts
The Instagram and LinkedIn algorithms reward time-on-content. Carousels deliberately consume more attention than single posts because users swipe through them.
April 2026 benchmarks (from agency reports and Buffer's 2026 social trends):
- LinkedIn carousels: ~3.5x the engagement of single-image posts in the same niche
- Instagram carousels: ~1.8x the engagement of single-image posts; higher save rate
- Re-share rate: carousels are saved at 4-7x the rate of single posts because they're often re-referenced
The narrative format matters. Carousels work as micro-stories. Single posts can't.
---
What gpt-image-2 Actually Does That's New
Here's what was hard before April 2026:
You'd write the carousel concept. A designer would draft 10 sketches. Then they'd generate or shoot 10 images. Then they'd grade them all to match. Then they'd compose typography per slide. Each slide an independent project, then a consistency pass.
Here's what changed:
A single prompt to gpt-image-2 generates 4-8 visually coherent images at once. Same character if there's a character. Same lighting style. Same color palette. Same aesthetic vocabulary. Built-in. Not "we manually checked."
The remaining work — typography, brand-color enforcement, headline copy — happens in Figma in 20-30 minutes. The whole carousel ships in under an hour.
---
The Six-Step Carousel Workflow
Step 1: Decide the Carousel Type
Three carousel types cover 90% of high-engagement posts:
- 1 title slide + 5-8 list slides + 1 CTA slide = 7-10 slides
- Each list slide: same composition, varying photographic content
- Highest LinkedIn engagement format
- 1 hook slide + 4-7 narrative slides + 1 CTA slide
- Each slide: progresses the visual story (same character, different scene/mood)
- Highest emotional resonance, highest save rate
- 1 title + 4-7 step slides + 1 CTA = 6-9 slides
- Each step: visual demo, screenshot, or process imagery
- Highest save rate; works for B2B education
Pick before you prompt. The structure determines the prompt.
Step 2: Write the Multi-Panel Prompt
The structure that works for gpt-image-2:
```
Generate a [N]-slide carousel for [platform: LinkedIn or Instagram].
Topic: [carousel topic in one sentence].
Aesthetic foundation [defined ONCE for the whole set]:
- Color palette: [3-5 hex codes or named colors]
- Lighting style: [natural / studio / golden hour / soft window light]
- Photography or illustration?: [pick one and commit]
- Mood: [3-5 adjectives]
- Visual reference: [one analogous brand or aesthetic — e.g., "Aesop campaign, Headspace illustrations"]
Per-slide content:
Slide 1 — [TITLE SLIDE]: [photographic/illustrative scene that supports
the carousel topic; leave clean space for headline overlay]
Slide 2 — [SECTION TITLE]: [scene that visually represents this section]
Slide 3 — [SECTION TITLE]: [scene that visually represents this section]
... [continue for all slides]
Slide [N] — [CTA SLIDE]: [composition with clean space for CTA text overlay]
Critical: hold visual consistency across all slides. Same color grading,
same lighting, same aesthetic vocabulary. Do NOT generate any text on
any slide — text overlays will be composited in Figma.
Use Thinking Mode for layout reasoning across the full set.
```
Step 3: Generate + Inspect for Consistency
Run the prompt. gpt-image-2 will produce all N slides in one generation (with Thinking Mode this takes 60-120 seconds).
Inspect for:
- Color consistency: does slide 4 have the same color grading as slide 1?
- Aesthetic drift: does any slide feel like it belongs to a different campaign?
- Character consistency (if applicable): does the character look like the same person across all slides?
If 80%+ of slides are coherent, work with what you have and patch the outliers in Figma. If less than 50% are coherent, regenerate.
The noise amplification bug applies here too — don't iterate the same prompt 5 times. After 2 retries, refine the prompt and start fresh.
Step 4: Compose Typography in Figma
Set up your carousel template in Figma:
- Create a frame at the platform's recommended dimensions:
- LinkedIn: 1200×1500 (portrait, 4:5 aspect)
- Instagram: 1080×1350 (portrait, 4:5 aspect) or 1080×1080 (square)
- Create N slide variants in the same Figma file
- Place each gpt-image-2 image into its slide
- Add headline + body copy per slide using your brand fonts
- Set up consistent text positioning across slides (Figma's auto-layout helps)
- Add slide numbers (e.g., "1/8") if your audience expects them
Branded carousels usually have a recognizable typographic system:
- LinkedIn: larger headlines, less body copy per slide
- Instagram: visual-first with shorter text overlays
- Both: consistent CTA styling on the final slide
Step 5: Multi-Format Export
The same Figma source exports to:
- LinkedIn (1200×1500)
- Instagram portrait (1080×1350)
- Instagram square (1080×1080) — optional secondary post
- Twitter (1200×675) — adapt the most powerful 2-3 slides as a thread
One generation, one composite, three platforms.
Step 6: A/B Test the Hook Slide
The hook slide (slide 1) determines whether anyone swipes to slide 2. Test 2-3 variants of the hook slide:
- Same imagery, different headline copy
- Same headline, different imagery
- Different aesthetic entirely
Run them as separate posts spaced 5-7 days apart. Track swipe-through rates in your platform analytics.
---
Three Production-Ready Prompts
Prompt 1: LinkedIn Listicle (8 Slides) — Founder Mistakes
```
Generate an 8-slide LinkedIn carousel.
Topic: "5 Mistakes Founders Make in Year One"
Aesthetic foundation:
- Color palette: deep navy (#1A2A3A), warm cream (#F1ECDF),
single muted gold accent (#C9A95C)
- Lighting style: soft natural window light
- Photography
- Mood: confident, contemplative, professional, warm
- Visual reference: editorial photography style similar to
Harvard Business Review or The Profile newsletter
Per-slide content:
Slide 1 (TITLE): empty modern desk with a laptop and coffee,
soft window light, blurred plant in foreground. Clean upper
third for title overlay.
Slide 2 (Mistake 1: Hiring Too Fast): empty meeting room with
several chairs, slight sense of absence.
Slide 3 (Mistake 2: Skipping Customer Calls): person at a desk
with laptop closed, looking at a phone. Photographic, slight
sense of avoidance.
Slide 4 (Mistake 3: Optimizing Vanity Metrics): laptop screen
with abstract dashboard imagery, slightly blurred so specific
numbers aren't legible.
Slide 5 (Mistake 4: Founder Isolation): solo founder at a desk
late evening, single lamp lighting. Cinematic, slight melancholy.
Slide 6 (Mistake 5: Pivoting Too Slowly): two paths in a forest
or two doors photograph, decision-moment imagery.
Slide 7 (TURNING POINT): same founder character from slide 5,
now in a daytime scene with another person, suggesting connection
and clarity.
Slide 8 (CTA): clean editorial composition with strong negative
space for the CTA text overlay.
Critical: hold consistent color grading across all 8 slides.
Same lighting style. Same editorial aesthetic. Do NOT generate
any text. All text composited in Figma.
Use Thinking Mode.
```
Prompt 2: Instagram Story Carousel (6 Slides) — Personal Brand
```
Generate a 6-slide Instagram carousel.
Topic: "How I Built a $1M Solo Business Without Hiring"
Aesthetic foundation:
- Color palette: warm cream (#F4EAE0), terracotta (#C97A4F),
deep teal (#2C5560), soft black (#1A1A1A)
- Lighting style: warm afternoon light, slightly cinematic
- Photography, illustrative-photographic hybrid
- Mood: optimistic, real, slightly nostalgic, achievable
Per-slide content:
Slide 1 (HOOK): first-person POV of hands typing on a laptop
at a coffee shop, single coffee cup beside. Soft afternoon light.
Clean upper area for hook headline.
Slide 2 (Year 1 — Struggle): same POV style, but at a messy
home desk, multiple coffee cups, papers, laptop. Slight chaos
photographic mood.
Slide 3 (The Decision): hand reaching for a notebook, single
pen, deliberate composition. Slight sense of resolve.
Slide 4 (Year 2 — Rhythm): person walking outdoors, golden hour,
back to camera. Sense of momentum.
Slide 5 (Year 3 — Rewards): same person on a balcony at evening,
laptop open but not the focus. Sense of arrival.
Slide 6 (CTA): clean composition, the laptop screen visible
but blurred, space for CTA text overlay.
Critical: hold cinematic warm aesthetic across all 6 slides.
Color grading consistent. Same person across all slides
(use character consistency feature). Do NOT generate any text.
Use Thinking Mode.
```
Prompt 3: B2B Tutorial Carousel (7 Slides) — SaaS
```
Generate a 7-slide LinkedIn carousel.
Topic: "How to Set Up a B2B Sales Pipeline in 5 Steps"
Aesthetic foundation:
- Color palette: clean cream (#F8F4ED), confident blue (#2B5F8B),
warm gray (#9C9389), single coral accent (#E07856)
- Lighting style: bright, clean, modern office natural light
- Photography with subtle illustrative overlays
- Mood: clear, professional, achievable, modern
Per-slide content:
Slide 1 (TITLE): clean modern desk, laptop with abstract dashboard
imagery (do not render specific data). Editorial composition.
Slide 2 (Step 1 — Define Your ICP): hands writing on a notebook,
abstract shapes representing customer segments.
Slide 3 (Step 2 — Build Your Lead List): laptop screen with
abstract list-imagery, blurred enough that specifics aren't
legible.
Slide 4 (Step 3 — Outreach Cadence): calendar imagery, clean
photographic style.
Slide 5 (Step 4 — Discovery Call Framework): two people in
conversation, professional but warm, soft natural light.
Slide 6 (Step 5 — Close + Follow-up): two hands shaking, modern
office context. Clear and professional.
Slide 7 (CTA): clean dashboard imagery (illustrative), space
for the CTA text.
Critical: hold a clean modern B2B aesthetic. Color palette
consistent. Photographic + illustrative blend held across slides.
Do NOT generate any text — all text in Figma.
Use Thinking Mode.
```
---
Three Carousel Mistakes That Kill Engagement
Mistake 1: Inconsistent Aesthetic Across Slides
If slide 4 doesn't visually match slide 1, the audience scrolls past. Multi-panel generation in gpt-image-2 holds aesthetic well, but you have to verify before publishing. The cost of catching this in Figma is 5 minutes. The cost of catching it after publishing is the engagement.
Mistake 2: Text in the Generated Image
Even though gpt-image-2 can render text, don't have it render your slide text. Two reasons:
- Your brand fonts are licensed; AI-rendered approximations of them are not
- Editing text means regenerating, which breaks visual consistency
Generate the imagery. Add text in Figma using your real brand fonts.
Mistake 3: Skipping the CTA Slide
Carousels without a clear CTA on the final slide convert at half the rate. Always design the last slide as a deliberate ask: follow, comment, save, click, subscribe, whatever your funnel goal is.
---
The Bottom Line
LinkedIn and Instagram carousels in April 2026 are the highest-engagement social format. Multi-panel coherence in gpt-image-2 is the production breakthrough that makes high-quality carousels economically rational for solo creators and small teams.
One prompt → 6-10 visually coherent slides → Figma composite in 30 minutes → ship. Compared to a half-day of designer time per carousel, this is the unlock.
But the rules apply: consistent aesthetic, brand fonts in Figma not in the generation, and a deliberate CTA on the last slide. Skip any of those and your carousel won't perform.
---
Get the Full Prompt Pack
ChatGPT Images 2.0 Prompts Pack includes carousel-specific prompts plus 25+ others, all weakness-aware. MIT-licensed, free.
For the foundational gpt-image-2 review: ChatGPT Images 2.0 Honest Guide.
For multi-panel storytelling beyond carousels: AI Storyboard + Comic Pack — 20 prompts for filmmakers, comic creators, and authors.
— Atilla