Igor Pogany tests ChatGPT Images 2.5 with reference portraits, repeated editsImage editing uses AI or conventional tools to change an existing picture while keeping the parts that should remain the same., object replacements and sketch-guided requests. He finds identity and edit continuity improved, but shows that changing a mug’s color also alters lighting and the table surface. A sketch resembling Homer Simpson is refused, illustrating a limit encountered during the demonstration.
Igor Pogany then describes a poster workflowWorkflow automation uses software or AI to complete a connected sequence of routine steps with less manual effort. that begins with two photos in Dropbox. Given a detailed brief, access to the necessary toolsAgent tool use is an AI agent's ability to select and operate external tools while completing a task. and an explicit stopping point, Astra creates and edits an image in ChatGPT, assembles the layout in Canva, exports a print-ready PDF and uploads it to an online printing service. Canva text remains editable, so the result can be revised after the automated work.
Igor Pogany reports that the process took about half an hour, including waiting and correction, and that Astra also stopped an OBS screen recording when finished. The poster was prepared for checkout but was not printed. He had already signed into the required applications and granted permissionHuman-in-the-loop command approval requires a person to review and authorize a consequential AI action before it executes., so the example does not demonstrate an entirely autonomous setup or a completed purchase.
Igor Pogany argues that the useful advance is preserving the same objective across several applicationsComputer use is an AI capability that interprets a graphical interface and operates software through actions such as clicking, typing, and selecting. while leaving intermediate files available for review and revision. He proposes video production and weekly operating reviews as further applications, but those are suggested extensions rather than completed demonstrations in this video.
Watch on YouTube
