We need to talk about the "Banana" in the room. For the last month, the AI image leaderboard has been dominated by a model with a ridiculous name: Nano Banana Pro. Google’s quiet release absolutely crushed the competition, leaving OpenAI’s DALL-E 3 looking like a relic from 2023. Nano Banana Pro Sam Altman reportedly declared a "Code Red." The team scrambled. And now, just days after dropping GPT-5.2, OpenAI has fired its response. It’s called GPT Image 1.5, and it’s rolling out to ChatGPT right now. GPT Image 1.5 Is it a Midjourney killer? Or just a panic patch to stop the bleeding? I spent the last 24 hours testing it, and the results are... complicated. Here is the no-BS breakdown for devs and creators. The "Studio" Shift: Chat is Dead The first thing you’ll notice isn’t the pixels; it’s the interface. interface For years, generating images meant typing draw a cat into a chat box and hoping for the best. It was clumsy. It was stateless. draw a cat OpenAI has finally realized that designers don't work in chat bubbles. designers don't work in chat bubbles. The new update adds a dedicated "Images" tab in the sidebar. It feels less like a chatbot and more like a simplified Photoshop or Canva. "Images" tab Preset Styles: No more memorizing 50-word prompts for "Cyberpunk Synthwave."
Aspect Ratio Toggles: Finally, native controls without typing --ar 16:9.
The "Inpainting" Grail: You can now highlight a section of an image and say "Make this coffee cup a beer," and it actually respects the lighting and shadows of the original scene. Preset Styles: No more memorizing 50-word prompts for "Cyberpunk Synthwave." Preset Styles: Aspect Ratio Toggles: Finally, native controls without typing --ar 16:9. Aspect Ratio Toggles: --ar 16:9 The "Inpainting" Grail: You can now highlight a section of an image and say "Make this coffee cup a beer," and it actually respects the lighting and shadows of the original scene. The "Inpainting" Grail: actually ![Image Description: A split screen comparison. Left: Old DALL-E 3 interface with a simple chat bubble. Right: The new 'Creative Studio' interface with sidebar tools, style presets, and a persistent canvas.] (Caption: OpenAI finally gave us a UI that respects the workflow.) 4x Speed & The "Consistency" Fix The biggest complaint with DALL-E 3 was that it was slow and hallucinatory. You’d ask to change a shirt color, and it would rewrite the user’s entire face. GPT Image 1.5 claims to solve this with Consistency Preservation. Consistency Preservation I tested this by uploading a photo of myself and asking it to put me in a spacesuit. Old Model: Generated a random astronaut who looked vaguely like my cousin.
GPT Image 1.5: Kept my face, my lighting, and my expression, but swapped the clothes. Old Model: Generated a random astronaut who looked vaguely like my cousin. Old Model: GPT Image 1.5: Kept my face, my lighting, and my expression, but swapped the clothes. GPT Image 1.5: It’s not perfect (skin textures can still look a bit "plastic" or "porcelain doll," a common RLHF artifact), but it’s a massive leap for practical use cases like e-commerce mockups or storyboarding. Oh, and it’s 4x faster. The "waiting for generation" spinner is almost gone. 4x faster The Benchmarks: OpenAI vs. Google vs. Midjourney Here is the spicy part. Is it actually #1? According to the LMArena leaderboard, GPT Image 1.5 instantly snatched the crown from Google’s Nano Banana Pro. But looking at the developer discussions on X and Reddit, the consensus is split: LMArena Realism: Google’s Nano Banana still wins. It captures "messy" reality—skin pores, bad lighting, ISO noise—better. OpenAI’s images still have that glossy "I was made by a computer" sheen.
Text Rendering: OpenAI takes the W here. If you need a sign that says "Welcome to the HackerNoon Annual BBQ", GPT Image 1.5 spells it correctly 95% of the time. Google still struggles with typos.
Censorship: This is where OpenAI loses. The "Safety" guardrails are incredibly aggressive. Trying to generate anything remotely "edgy" or copyrighted triggers the dreaded "I cannot generate that" refusal. Google and open-weights models are far more lenient. Realism: Google’s Nano Banana still wins. It captures "messy" reality—skin pores, bad lighting, ISO noise—better. OpenAI’s images still have that glossy "I was made by a computer" sheen. Realism: Text Rendering: OpenAI takes the W here. If you need a sign that says "Welcome to the HackerNoon Annual BBQ", GPT Image 1.5 spells it correctly 95% of the time. Google still struggles with typos. Text Rendering: "Welcome to the HackerNoon Annual BBQ" Censorship: This is where OpenAI loses. The "Safety" guardrails are incredibly aggressive. Trying to generate anything remotely "edgy" or copyrighted triggers the dreaded "I cannot generate that" refusal. Google and open-weights models are far more lenient. Censorship: "It’s better than Nano Banana, but MUCH worse than Nano Banana Pro." — Top comment on Hacker News "It’s better than Nano Banana, but MUCH worse than Nano Banana Pro." — Top comment on Hacker News "It’s better than Nano Banana, but MUCH worse than Nano Banana Pro." For the Devs: API Price Drop If you’re building apps on top of this, the news is good. OpenAI has slashed the API pricing for image generation by 20% with the gpt-image-1.5 endpoint. This is a direct shot at Stability AI and Midjourney (which still lacks a public API). By making high-fidelity, editable image generation cheaper, OpenAI is trying to lock in the developer ecosystem before Google’s Gemini API takes over. The killer feature for devs? Structured editing. You can now pass an image + a mask + a prompt via API, effectively giving you programmatic Photoshop. The killer feature for devs? ![Image Description: A bar chart showing 'Time to Generate'. The bar for 'DALL-E 3' is tall (15s), while 'GPT Image 1.5' is tiny (3s). A second chart shows 'Cost per 1k Images' dropping by 20%.] (Caption: Faster, cheaper, but is it better?) The Verdict: A "Good Enough" Killer? OpenAI isn't trying to beat Midjourney on art. They are trying to beat Canva on workflow. Canva GPT Image 1.5 isn't the most artistic model in the world. It won’t win digital art competitions. But it is the most usable model. The combination of the new UI, the text rendering, and the speed makes it the "Good Enough" tool that 90% of users will default to. usable Google’s Nano Banana might be the connoisseur's choice, but OpenAI just built the iPhone camera of AI art: Point, shoot, and it works. 5 Takeaways for Techies: The UI is the Product: The shift to a "Creative Studio" sidebar is more important than the model upgrade.
Consistency is King: The ability to edit parts of an image without rerolling the whole seed is the feature to watch.
Price War: 20% cheaper API costs means we’re about to see a flood of new AI-gen apps.
Google is Still Scary: The fact that OpenAI had to rush this out proves they are terrified of Gemini/Nano Banana.
Text Rendering: Finally usable for memes, logos, and UI mockups. The UI is the Product: The shift to a "Creative Studio" sidebar is more important than the model upgrade. The UI is the Product: Consistency is King: The ability to edit parts of an image without rerolling the whole seed is the feature to watch. Consistency is King: parts Price War: 20% cheaper API costs means we’re about to see a flood of new AI-gen apps. Price War: Google is Still Scary: The fact that OpenAI had to rush this out proves they are terrified of Gemini/Nano Banana. Google is Still Scary: Text Rendering: Finally usable for memes, logos, and UI mockups. Text Rendering: Liked this breakdown? Smash that clap button and follow me for more deep dives into the AI wars. Liked this breakdown? Smash that clap button and follow me for more deep dives into the AI wars.

Render

Google

OpenAI Just Hit "Code Red": Inside the New ChatGPT Images Update

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Google Just Killed LeetCode: Why AlphaEvolve is the End of the "Algorithm Interview"

The Noonification: Use This 7-Step McKinsey Framework to Solve Any Problem (1/10/2023)

The Noonification: A Taxonomy of Inclusiveness (1/11/2024)

The Noonification: What is the InfiniteNature-Zero AI Model? (11/19/2022)

10 Ways AI Has Changed Our Lives

100 Days of AI, Day 8: Experimenting With Microsoft's Semantic Kernel Using GPT-4

Google Just Killed LeetCode: Why AlphaEvolve is the End of the "Algorithm Interview"

The Noonification: Use This 7-Step McKinsey Framework to Solve Any Problem (1/10/2023)

The Noonification: A Taxonomy of Inclusiveness (1/11/2024)

The Noonification: What is the InfiniteNature-Zero AI Model? (11/19/2022)

10 Ways AI Has Changed Our Lives

100 Days of AI, Day 8: Experimenting With Microsoft's Semantic Kernel Using GPT-4

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps