How to Upscale a Photo to 8K in ChatGPT Without a Separate Upscaler

To increase the resolution of a photo, people usually take a separate upscaler: Topaz, Gigapixel, an online service with a limit of free attempts. In early October 2026, one niche AI channel came across a different approach: the same result is achieved right in ChatGPT, with no program installs and no separate upscaler subscription.

What the Work mode in ChatGPT is

On July 9, 2026, OpenAI released ChatGPT Work: a mode in which the model works not as a conversation partner but as an agent. The model splits a big task into stages and makes intermediate decisions itself. It can run one process for hours with almost no human involvement. On desktop the agent additionally gets Computer Use: it can click buttons, type text, and work with files directly on the computer.

At release the mode was available on the Pro, Enterprise, and Edu plans, with a promise to roll it out to Plus and Business in the coming days. For photo upscaling, this is the key difference from an ordinary chat: you can describe the whole processing procedure to the agent once, instead of sending commands for every step.

How the method itself works

The idea is simple and honest, with no magic. It is hard for an agent to pull one large image up to high resolution as a whole: detail gets blurred and the model loses small elements. The method solves this by splitting into parts.

The photo is divided into 9 overlapping pieces on a 3x3 grid. The agent improves each piece separately, with margin on the edges so there is no visible seam at the joint. After processing, all 9 parts are put back together into one image by aligning on their shared borders.

Then the method repeats the same trick once more, but no longer across the whole frame, only across the most important zones such as the face, hands, clothing texture, and small details of objects. On the third pass the focus narrows once more, down to individual elements of the face and the texture of materials. Each pass preserves proportions, facial expression, the natural imperfections of the skin, and the direction of the fabric, without touching what already came out right. The author of the method separately tells the agent what it must not do: not invent details, not add a glow effect, not smooth skin into plastic, not change the color and shape of objects.

What you end up with

At the output the agent assembles the final image at the set resolution, 8K on the long side by default, and saves it as an uncompressed PNG. It also states directly that some details look not like the original pixels but like a plausibly reconstructed texture: skin, hair, and fabric are drawn anew based on the overall logic of the image, not read from an original that physically wasn't there.

This is the honest limit of the method. Upscaling through ChatGPT doesn't restore information that wasn't in the source file. It convincingly fills it in, and the higher the original quality of the photo, the less the agent has to guess.

When the method is worth it and when an ordinary upscaler is simpler

The method makes sense if you already have access to ChatGPT Work and don't want to pay for a separate service for a one-off task: enlarging a photo for print, a cover, a portfolio. The process takes a few minutes and requires no program download.

If you need to process dozens of photos in a batch, or if maximum restoration accuracy matters (for example, for forensic examination or scientific photography), a dedicated upscaler with a specialized model remains more accurate. ChatGPT Work suits single shots where you need a beautiful, convincing result, not metrological precision.

What to do before trying it yourself

Before running it, decide two parameters in advance: how many detail passes you need (two are usually enough, the third adds processing time but not always a difference visible to the eye) and what final resolution is really required. Pushing a photo to 8K for posting on Instagram makes no sense: they will compress the picture on delivery anyway. 8K is justified for large-format print or for working with details of the frame.

I write about which AI combinations I really use in my work with visuals and content in the channel «Растём с ИИ» ("Growing with AI"). If you need help setting up such automation for your task, you can start with a free consultation.