Yesterday, OpenAI shipped GPT Image 2. It renders text at 99% accuracy. It does 4K. It took the top of the Image Arena leaderboard with a 242-point Elo lead over every other model. Infographics, slides, ad creative, quote cards with typography that actually looks designed — all from one prompt, in a chat window you already have open.
Creation, for most people, is now solved.
Here is what I did with it this morning. I prompted it for a carousel cover. It came out beautifully. Then I sat there, looking at the image, realising I still had to: download it, open Instagram, caption it, open LinkedIn, reformat the caption, re-upload the image, open TikTok, add the AI label, open X, trim the caption to 280, open Bluesky — you know this list. You have written it a hundred times.
Creation is solved. Distribution is still, somehow, manual.
The gap ChatGPT cannot close
OpenAI is not going to build a social media scheduler. Their job is to make the model better, hold you inside ChatGPT, and let you export. The calendar, the per-platform formatting, the AI-content labels, the way every platform wants your images sized differently, the analytics loop — none of that is where their incentives point.
So you get this strange moment. The hardest, most expensive part of content — the creative itself — costs you a sentence and a few seconds. The easiest, most mechanical part — moving a file from point A to point B — still takes the same fifteen minutes it did in 2018.
Every creator hits this wall this week. They have a brilliant image sitting in a ChatGPT tab, and no way to get it live.
Distribution is the actual product
I have been building Rheos for eighteen months. For most of that time I would have told you it was an AI content tool. That framing was wrong, or at least incomplete. What Rheos actually is — and what I think every AI creation tool should be paired with right now — is a distribution layer.
Brand voice, platform formatting, image sizing per channel, scheduling, the AI-content labels, the hundred small things each platform wants you to do differently. That is the unglamorous middle. It is also the part that decides whether your beautiful GPT Image 2 output ever touches a single human on Tuesday morning.
I realised how true this was about an hour ago. I am testing the thing that lets Claude publish directly from the chat. The test I am running right now: draft this exact article, generate a cover image with Nano Banana, schedule it to LinkedIn and X, and push the long-form to Medium and Substack. From a chat window. Without touching our dashboard once.
If that works, and I think it will today, the whole model of content creator shifts. You don't sit in the app. You sit in Claude, or ChatGPT, or whatever model you prefer, and you say post this. The tool that owns the last mile — the distribution — wins.
What the last mile actually looks like
The reason nobody talks about distribution is that it is boring. Let me make it slightly less boring by listing what actually happens between a GPT Image 2 creation and a scheduled post on nine platforms:
- Resizing the image for Instagram square, Pinterest vertical, X landscape, LinkedIn banner
- Writing a caption that fits Instagram's 2,200-character style, rewriting it longer and more professional for LinkedIn, trimming it to 280 for X
- Ticking the AI-generated-content box at upload on TikTok, Instagram, and YouTube — because if you don't, TikTok alone will quietly throttle your reach by up to 50%
- Scheduling each post at the right time for its platform — LinkedIn at 08:00 UK, Instagram at 18:30, Pinterest whenever, TikTok mid-evening
- Staying connected to every platform so the publish doesn't silently fail overnight
- Tracking what went where so you know what worked and what to repost
Every AI tool generates. One or two tools close this loop. That, I think, is the real category this year.
Three predictions
-
By June, every serious scheduler will have an AI-image integration. They will all be chasing GPT Image 2. The ones that win will be the ones that treat the image as the start of the workflow, not a thing uploaded from a phone.
-
You stop opening dashboards. You ask the model to post. The interface becomes the chat window — Claude, ChatGPT, whatever you prefer. Whoever owns that publish step becomes the Buffer of this decade.
-
The "how do I post this everywhere" question becomes the most-searched social-media query of the next 90 days. GPT Image 2 is the reason. Gemini's next model will add fuel. Every agency, every SME, every creator has the same question and no clean answer right now.
The end
The entire social media industrial complex spent ten years optimising the easy part — creation — while the hard part, the move from create to published-across-nine-platforms, stayed mostly manual. GPT Image 2 just lapped that finish line. The creation problem is, for all intents and purposes, solved.
The distribution problem is wide open. It is not glamorous. It does not trend on X. But it is the last real moat in AI content, and the next year is going to be a race for it.
I am publishing this article from a Claude chat window, straight into Medium, Substack, and LinkedIn, to test exactly that. If you are reading this today, you can see how it went.