A still that goes out with a company release used to mean a reshoot when one label, one line of type, or one prop was wrong. The page for the next image editor describes a narrower change. You lay a mark on the picture, say what should change, and leave the rest. That editor is named. It is not switched on. The brief for the look can still be kept while the team waits, and while they correct the still with the editor that is actually open.
This is not a ranking of video makers. No platform was screened against another. The two pages below are the ones this piece can check.
1. The mark on the still
GPT Image 2.6 is named in OpenAI’s image-edit spec. The page says it has no public model ID, no release date, and no price. On Topview the control is marked coming soon. The page’s own answer, dated 30 October 2026, is that 2.6 is not in the model list, the API changelog, or the image generation guide. Checked against that same page, it is still not connected. The frames on the page are a markup workflow run with GPT Image 2.5. The page says they are not official 2.6 samples, because 2.6 is not released. What you can open on Topview today is GPT Image 2.5 Sunburst. This piece does not add a second link for that live editor.
What the spec does describe is markup. A transparent PNG is laid over the first image to guide the edit, in place of a mask. The overlay has to match the first image’s size after orientation, and it has to contain transparent pixels, so the marks sit where you put them. One request uses markup or a mask, not both. OpenAI has not said how the model reads each mark. Arrows, circles, and handwriting are not a documented language on this page.
The page lists the jobs that markup is for, with the live editor until 2.6 arrives.
A master visual for more than one market. Mark the copy and the sign areas, and ask for that market’s text. A fluent reviewer still checks every language. A marked layer does not make a translation correct.
A product still. Mark a label, a cap, or a prop, and bring the real packaging as a reference if you have the right to use it. The rest of the shot is what you asked to keep. Small type can still come back wrong. If the release has to show the real package, check it against the real package.
A keyframe. Fix a prop, an outfit, or a background on the still so a later video would start from a frame someone approved. The page says that approved still can be sent on to image-to-video. It also says GPT Image 2.6 is not connected to that step.
A review note. Mark the change on the frame and add the note. The comment stays on the pixels it is about. Versions can sit on the Board for the team to look at.
The steps on the page are short. Add the image from the Board, and add a reference such as a logo or a style frame if you need one. Mark the part that should change on a transparent layer the same size as the image. Say what should change and what must stay. Then stop, if 2.6 is the tool you were waiting for. The button is not live.
The page does not publish a 2.6 price, an output size, or a date. Limits printed beside the spec, including how many images an edit request can hold and how long a prompt may be, are labeled as the current edit endpoint, not as a 2.6 specification.
2. The brief that should still be there for the next version
While the still is in review, the same notes get typed again. The market. The product. What must not move. What the last version got wrong.
A Topview Bot sits in the Board and can remember preferences such as lighting, color, and pacing. It can bring those preferences from ChatGPT, from Claude, or from a Board you already use. You can also say the rules in a sentence. Read what it kept. A look saved for a dark hero image will miss a bright package shot.
On the Board, the conversation and the generation stay in one place. A supported asset can go into the next task without a download and a second upload. Versions of the still, the references, and the notes can be grouped by scene, by stage, or by purpose. Ask for that grouping. Check that this release did not land in another project’s pile.
Memory is the look and the working habits. A Skill is a sequence you want to run again, such as gathering references, planning the frames, and producing the assets. The page does not ship a skill named for an earnings still or a product release. You save the steps you repeat.
The Bot is not the image editor, and it is not a video model. It does not switch GPT Image 2.6 on. Supported shots can generate at the same time when the model and the account allow it. That is a later step, after the still is one you accept. Sharing a configured Bot with a colleague is marked as a preview. What travels with it, who can open it, and when that share is finished can still change. The Bot page does not print a separate price for the Bot.
What the release still has to get right
Mark the master. Have a fluent person read every market’s words. Keep the real package beside the still if the label matters. Let the Bot hold the rules so the next version does not start from a blank note. Leave the video until the frame is approved, and do not wait on 2.6 for that video. It is not connected.
A marked edit does not make a claim in the release true. A generated room is not the plant, the store, or the event. People in the frame are characters in the picture unless they agreed to appear. Anything you upload has to be something you may use. Cleaning a wrong fact out of the caption is an editing job. It is not something the markup layer does by itself.
The professional step on these two pages is small on purpose. One of them names an editor you cannot press yet, and says what the mark is supposed to do when you can. The other remembers the brief, groups the versions, and stays inside the Board you already have.