L
Liora
Guest
I asked Codex to use Photopea to turn a product-style photograph into a blog cover. The task involved opening a browser editor, arranging an image and headline, keeping the composition editable, and delivering files I could use. Codex handled the editing steps; I stepped in to confirm the browser download.
The session ran GPT-6 Astra, confirmed in its task record. In its Astra announcement, OpenAI describes improvements to computer use and updates to the Codex system that support those interactions. This small exercise gave me a concrete example to inspect: a 1600 × 900 JPG cover and a PSD master made from a public-domain coffee photograph. It was not a comparison with an earlier model, so it cannot establish how much faster Astra made this task.
Photo by mcfoodie, via Wikimedia Commons (CC0). Cover layout created in Photopea.
Photopea was the editing workspace, and Codex operated its interface. It selected menus, entered dimensions, arranged layers, and inspected the changing screen. The image editing happened through Photopea's ordinary controls, without a paid image-generation step. Codex also used local file checks to verify the downloaded result.
That combination is useful for a creator who knows the desired deliverable but does not want to manage every editor command. The request still needs direction: the output size, the photo to use, room for a headline, and an editable master. The person requesting it still decides whether the finished design is worth publishing. A basic layout exercise is a good place to see where that division of work helps.
The working canvas was 1600 × 900 pixels, a 16:9 layout. This is a design choice for the exercise. Check the destination’s current requirements before preparing a final upload: a website header, a social preview, and a card thumbnail may show different portions of the same image.
The source was Coffee cup and coffee bean, credited to mcfoodie and marked CC0 on its individual Wikimedia Commons page. It is a practice still life, not a photograph of a product sold or tested by Liora. The original measures 1547 × 1024 pixels.
The source photograph by mcfoodie, via Wikimedia Commons (CC0), before the Photopea layout.
Leave a visible margin around the object and keep the title away from the outside edges. These are composition decisions to check in the publisher’s preview, rather than guarantees that every automated crop will work. If the card preview loses the title, prepare a separate version with less text or none.
Photopea distinguishes changing image dimensions from changing the canvas window. That distinction matters when making a wide cover: expanding the canvas creates room around the composition; stretching the whole photograph changes the product's proportions. Codex used those controls in sequence during this exercise.
Codex opened the photo through the local file picker. It used Image → Image Size to reduce it proportionally to 900 × 596 pixels, then Image → Canvas Size to expand the composition to 1600 × 900, with the anchor set to the middle-right. This left the original photo background intact and created space on the left, above, and below it.
One detail needed an extra step: the expanded area was transparent. Codex added a solid-color fill layer, set it to #f5f1e9, and moved it below the photograph. The warm background now filled the extra space. This is a useful part of computer use: inspecting the intermediate state and continuing with another operation when the canvas is not yet ready.
The original photograph already cuts into the cup on its right edge; expanding the canvas cannot restore that missing part. For a commercial cover that needs the whole object, choose a less tightly framed source.
This small exercise resized a raster photograph once. For a layout likely to need repeated scaling, a Smart Object would retain the source data for later revisions. That would be an improvement to request on a subsequent run; it was not the method used here.
Codex placed “A Better Product Cover” across three lines on the left, using 70-pixel DejaVu Sans Book text on a separate layer. This short phrase fits the practice image; a published cover should use wording that matches its article.
The layer structure matters as much as the preview. With separate photo, text, and background layers, the headline can be revised without rebuilding the composition. The saved layer-panel screenshot showed those three pieces separately.
Judge the arrangement at a small display size as well as at full size. If the product becomes a vague shape or the title needs zooming to read, reduce the number of competing elements. There is little value in adding a subtitle that disappears in the place where readers first encounter the cover.
Photopea’s save documentation distinguishes an editable PSD from a web export. Both were saved for this exercise. The JPG used 1600 × 900 pixels at 70% quality. The downloaded file measured 97,819 bytes, matching the size shown in the export dialog.
The download was also where I needed to help. The browser's automated download event timed out, and I manually confirmed the download. Codex then checked the file on disk and reopened the JPG to inspect the headline, background, and photo placement. The three title lines remained visible, and the cup's existing crop matched the source. The PSD was saved and its file header checked, but it was not reopened for a second layer-by-layer inspection.
For me, the useful result of Astra-powered Codex here was a completed sequence of editor operations plus files that could be inspected afterward. The cover is intentionally simple. It does not demonstrate professional art direction or prove that every browser task can run unattended. It does show a practical way to delegate a small creative production task while keeping human judgment at the points that matter: the brief, the handoff, and the final design.
The session ran GPT-6 Astra, confirmed in its task record. In its Astra announcement, OpenAI describes improvements to computer use and updates to the Codex system that support those interactions. This small exercise gave me a concrete example to inspect: a 1600 × 900 JPG cover and a PSD master made from a public-domain coffee photograph. It was not a comparison with an earlier model, so it cannot establish how much faster Astra made this task.
Photo by mcfoodie, via Wikimedia Commons (CC0). Cover layout created in Photopea.
What computer use meant in this task
Photopea was the editing workspace, and Codex operated its interface. It selected menus, entered dimensions, arranged layers, and inspected the changing screen. The image editing happened through Photopea's ordinary controls, without a paid image-generation step. Codex also used local file checks to verify the downloaded result.
That combination is useful for a creator who knows the desired deliverable but does not want to manage every editor command. The request still needs direction: the output size, the photo to use, room for a headline, and an editable master. The person requesting it still decides whether the finished design is worth publishing. A basic layout exercise is a good place to see where that division of work helps.
Choose the frame before arranging the photo.
The working canvas was 1600 × 900 pixels, a 16:9 layout. This is a design choice for the exercise. Check the destination’s current requirements before preparing a final upload: a website header, a social preview, and a card thumbnail may show different portions of the same image.
The source was Coffee cup and coffee bean, credited to mcfoodie and marked CC0 on its individual Wikimedia Commons page. It is a practice still life, not a photograph of a product sold or tested by Liora. The original measures 1547 × 1024 pixels.
The source photograph by mcfoodie, via Wikimedia Commons (CC0), before the Photopea layout.
Leave a visible margin around the object and keep the title away from the outside edges. These are composition decisions to check in the publisher’s preview, rather than guarantees that every automated crop will work. If the card preview loses the title, prepare a separate version with less text or none.
Photopea distinguishes changing image dimensions from changing the canvas window. That distinction matters when making a wide cover: expanding the canvas creates room around the composition; stretching the whole photograph changes the product's proportions. Codex used those controls in sequence during this exercise.
Place the product without distorting it.
Codex opened the photo through the local file picker. It used Image → Image Size to reduce it proportionally to 900 × 596 pixels, then Image → Canvas Size to expand the composition to 1600 × 900, with the anchor set to the middle-right. This left the original photo background intact and created space on the left, above, and below it.
One detail needed an extra step: the expanded area was transparent. Codex added a solid-color fill layer, set it to #f5f1e9, and moved it below the photograph. The warm background now filled the extra space. This is a useful part of computer use: inspecting the intermediate state and continuing with another operation when the canvas is not yet ready.
The original photograph already cuts into the cup on its right edge; expanding the canvas cannot restore that missing part. For a commercial cover that needs the whole object, choose a less tightly framed source.
This small exercise resized a raster photograph once. For a layout likely to need repeated scaling, a Smart Object would retain the source data for later revisions. That would be an improvement to request on a subsequent run; it was not the method used here.
Give the headline its own space.
Codex placed “A Better Product Cover” across three lines on the left, using 70-pixel DejaVu Sans Book text on a separate layer. This short phrase fits the practice image; a published cover should use wording that matches its article.
The layer structure matters as much as the preview. With separate photo, text, and background layers, the headline can be revised without rebuilding the composition. The saved layer-panel screenshot showed those three pieces separately.
Judge the arrangement at a small display size as well as at full size. If the product becomes a vague shape or the title needs zooming to read, reduce the number of competing elements. There is little value in adding a subtitle that disappears in the place where readers first encounter the cover.
Save the composition and inspect the export.
Photopea’s save documentation distinguishes an editable PSD from a web export. Both were saved for this exercise. The JPG used 1600 × 900 pixels at 70% quality. The downloaded file measured 97,819 bytes, matching the size shown in the export dialog.
The download was also where I needed to help. The browser's automated download event timed out, and I manually confirmed the download. Codex then checked the file on disk and reopened the JPG to inspect the headline, background, and photo placement. The three title lines remained visible, and the cup's existing crop matched the source. The PSD was saved and its file header checked, but it was not reopened for a second layer-by-layer inspection.
For me, the useful result of Astra-powered Codex here was a completed sequence of editor operations plus files that could be inspected afterward. The cover is intentionally simple. It does not demonstrate professional art direction or prove that every browser task can run unattended. It does show a practical way to delegate a small creative production task while keeping human judgment at the points that matter: the brief, the handoff, and the final design.