Photoshop Inside ChatGPT: What Works, What Breaks

Using Adobe Photoshop inside ChatGPT changes the math on quick edits, especially when the job is small but time is tight. The catch is that the results can look polished one minute and sloppy the next, so knowing what it can and cannot do saves missed deadlines and rework.

Coming to you from Anthony Morganti, this practical video walks through how the Photoshop connection inside ChatGPT actually works, step by step, without assuming a paid account. Morganti starts where most people get stuck: finding the app connection in settings, connecting it once, then selecting Photoshop from the “more” menu before uploading an image. He points out a file limitation that will matter immediately when your workflow is raw-first: the editing inside ChatGPT is effectively limited to JPEGs in his testing, with other formats triggering locked-feature messages. He also shows what “talking” an edit into existence looks like in real-time, including the waiting, the previewing, and the way the interface presents multiple interpretations of a vague prompt. If you have been wondering whether this is a real tool or a demo gimmick, the on-screen sequence answers that fast.

The most useful part is how clearly the video shows the difference between edits that are mostly global and edits that rely on precise masking. A “vintage” request produces multiple options, like a sepia-style treatment and a grain-only look, plus sliders to tune highlights, shadows, and contrast, and even a midtone color picker. Then Morganti moves into targeted color changes, like shifting the hue of a robe, and you get a better sense of what the tool is doing behind the scenes: it tries to isolate a region, then hands you a control to steer the hue. When it works, it is quick and surprisingly clean. When it misses, it misses in a way you will recognize, bleeding into skin or hair because the selection is broader than it claims. The video also shows a quiet skill that matters more than any button: writing prompts that describe the exact area, not the concept of the edit.

Morganti pushes the tool with a harder request, blurring a background in a composite where the separation between subject and environment is not obvious. You see the “hit or miss” reality, including how the same image can behave differently across attempts, and how adding specificity can rescue the outcome. After that, he shows a failure mode that should shape how you use this: a request to make a dress blue that also shifts skin tone and hair, even while the tool claims it will not. From there, Morganti demonstrates the escape hatch, opening the file in Photoshop on the web to manually fix the mask with brushes and quick edits. The limitation is operational, not artistic: you cannot save that web edit back into ChatGPT, so the loop becomes download, then re-upload, then continue from the new version.

He closes with a blunt opinion on where AI fits in image-making, and it is not a cheerleading segment. Morganti argues that constraints build craft, using a teaching exercise that forces you to shoot with a simple setup and commit to decisions in-camera, then contrasts that with AI as a box of near-unlimited options that can replace the hard part with prompts. He also calls out the social pressure around AI, including creators who celebrate expensive tiers and treat the tooling like a status signal, and he invites disagreement rather than trying to win the room. If you already own and know Adobe Photoshop, his advice is pointed: do serious work where you have full control, then treat the ChatGPT integration as a convenience layer when speed matters more than perfection. Check out the video above for the full rundown from Morganti.