Copilot Generates Images. Don't Ship Those as Product Stills.

GitHub’s changelog on 28 October 2026 says Copilot can now generate and edit images from natural language. The selected reasoning model calls a dedicated image tool powered by GPT Image 2.5 Flare, previews the result inline, and saves it into the project. That file is an in-repo concept. It is not MAIN, not an App Store screenshot, not a signed ad, and not proof the SKU exists. Lock catalog and campaign stills from a photograph you already trust. Leave type empty. Overlay real words last. FreeGPTImg is a GPT Image 2 workspace: 20 daily free credits, no credit card required. Do not stamp Flare on a GPT Image 2 JPEG, and do not ship a Copilot chat still as the product.

Draft in FreeGPTImg. Stay on GPT Image 2 when that is the label on the screen. Keep layout-only UI concepts in the AI UI generator. Recrop a feed frame in AI social media ads. Keep materials honest in AI product photography. Cut a catalog crop in the Amazon product photo generator. Change only one noun from a locked hero in image to image. Start a net-new concept in text to image only when you do not yet have a photo.

What the job actually is

A PNG that landed in docs/ because you asked Copilot for “a hero shot of our kettle” is a conversation artifact. The kettle you ship is still the kettle you photographed. If the Copilot still grows a new spout, a fake rating pill, an invented OS status bar, or a “Made with AI” watermark you then upload as MAIN, you mixed two jobs.

OpenAI’s current image guide still splits the work: generate from a prompt, or edit an existing image, including partial edits. The same guide currently points API users at gpt-image-2.5-flare for everyday generation and gpt-image-2.5-sunburst when editing precision matters, with xhigh and max quality tiers above high. Background can be transparent, opaque, or auto. For gpt-image-2, quality stops at low / medium / high, and you omit input_fidelity because every image input is already processed at high fidelity. Copilot’s changelog names Flare for the in-chat tool. FreeGPTImg currently exposes GPT Image 2 on this loop. Write the name on the screen into the filename.

GitHub’s own limits matter. Image generation is in public preview for Copilot Pro, Copilot Pro+, Copilot Business, and Copilot Enterprise. It is billed separately from the chat model, on image tokens that count toward the plan’s AI credits. Organization and enterprise owners can turn the image model off in Copilot model policies. You can opt in to a visible “Made with AI” watermark; it is off by default. Recheck the live changelog. None of that makes the file a product photo.

Amazon’s product image help still treats MAIN as a realistic photo of the item you sell, on a pure white field (RGB 255, 255, 255), without extra graphics or type. A README diagram and a Copilot mock of a “listing” do not replace MAIN. Meta’s image-ad guide for Facebook Feed currently recommends JPG or PNG, a 4:5 frame, and 1440×1800 as a working resolution, with a 600 px minimum width. That is a crop target, not permission to bake a Copilot CTA into pixels.

The cheap failure this week is not “I did not try Copilot.” It is “the model invented a cousin bottle in chat, I committed it, and now the storefront disagrees with the box.”

Name the still before you generate

Weak sets fail because one Copilot prompt tries to be a README diagram, an App Store screenshot, MAIN, a 4:5 ad, and the signed headline. Pick one job.

Job	What you export	Keep	Change one thing	Wrong move
In-repo concept still	A README diagram, mood board, or placeholder	One idea	Nothing else	Uploading it as the product photo
UI mock still	Layout only, empty copy	Frame, spacing	Empty type	Fake ratings, fake OS chrome, invented features
Locked catalog still	The SKU you already ship	Shape, color, label, logo, pack size	Nothing, or light only	Starting from the Copilot JPEG
White MAIN draft	Same SKU, square, opaque white	The upload	Field only	Watermark, UI chrome, extra unit
4:5 feed still	Same SKU, empty lower or top band	Silhouette, palette	Crop and empty band	Baking the Copilot CTA
9:16 story still	Same SKU, empty lower third	The object	Crop only	Status bars, prices, “Made with AI”
Logo / mark concept	Simple raster mark	Silhouette	Nothing	Copilot watermark on the mark
Offer or UI overlay	Don’t generate it	Empty type	Editor, last	“4.9★”, “$29”, “Download” in chat

A text-to-image pass invents. An image-to-image pass should borrow. If you already paid for a pack shot that matches the listing, that file is truth. A Copilot chat does not get to grow a second lid.

Checklist before anything leaves the repo

Use this sequence. Do not paste a sprint ticket into Copilot and expect the JPEG to be the SKU.

1. Name the job. In-repo concept, UI mock, catalog still, MAIN draft, or ad still. One file, one job.
2. If it is a catalog or ad still, start from a photograph. The locked JPEG of the item you sell. If you do not have that file, stop and photograph it. Do not invent a bottle because Copilot is now in the sidebar.
3. Write keep / change / do-not in five lines. Keep silhouette, color, label, logo, and pack size. Change only one noun: crop, field, or empty band. Do not add a second SKU, a rating pill, a fake notch, or a percent.
4. Name the engine on the screen. If Copilot ran Flare, write copilot-flare into that filename and keep the file in docs/ or scratch/. If FreeGPTImg shows GPT Image 2, write gpt-image-2. Do not relabel either file sunburst-approved.
5. Leave the watermark off for anything that might ship. GitHub’s “Made with AI” mark is opt-in and off by default. Do not turn it on for catalog, MAIN, or ads. If a platform requires AI disclosure, follow that policy in copy or metadata, not as a baked badge unless you meant to.
6. Ask for empty type. Prices, percents, star ratings, “Download on the App Store,” legal lines, and model names never belong in the generator if they must be exact. Overlay them in an editor after a person signs the words.
7. Leave production room. Product or UI in the middle third. Quiet field. Empty band where the headline will sit. Do not fill every edge with chat chrome.
8. Generate one still. One pass. File names like kettle-4x5-empty-band-gpt-image-2.png. Never kettle-final-copilot.png for a listing.
9. Inspect the subject before the overlay. A new spout, floating “GPT-4”, a fake home indicator, a ribbon that says “SALE”, or a tag that says “4.9” are stop conditions. Hold the still next to the real pack or the real build. If a stranger would think this brand shipped a Copilot edition, discard it.
10. Only then overlay. In an editor, set the real headline and only the claims a manager signed. Recrop to the live placement size after the pixels match. Do not ask Copilot to spell the price.

Prompt formula for the locked still

GPT Image 2 catalog or ad still, not a Copilot chat still and not MAIN unless named.
Start from the uploaded product photo.
Keep: exact silhouette, color, label, logo, pack size.
Change only: [crop / field / empty headline band].
Empty type. No prices, percents, ratings, model names, watermarks, or store chrome.
Do not add: extra SKUs, fake OS bars, Copilot UI, “Made with AI”, celebrity faces, children.

Prompt examples

In-repo concept, clearly labeled. “Text-to-image: a simple architecture diagram of a kettle firmware loop, white field, boxes only, empty labels for later overlay. Not a product photo, not a store listing.”

UI mock, empty copy. “Text-to-image: a phone-sized habit-tracker frame, three empty rows, no status bar, no carrier, no battery percent, no star rating. Empty type. Not an App Store screenshot.”

4:5 feed, empty lower band. “Image-to-image: keep this exact matte-black kettle, spout curve, and handle. 4:5 crop, product in the upper two thirds, empty lower band for later overlay. Quiet kitchen, even light. No prices, no Copilot chrome, no second kettle.”

White MAIN, not a Copilot file. “Image-to-image: keep this exact SKU. Change only the field to opaque pure white. No badge, no watermark, no extra unit.”

Lifestyle in-use, type empty. “Image-to-image: keep this exact mug glaze and printed mark. Change only the scene to a quiet desk. No laptop showing Copilot chat, no price tag.”

Do not generate this. “App Store screenshot of our kettle app, 4.9 stars, $29.00, Download now, GPT Image 2.5 Flare inside Copilot, guaranteed Amazon approved, Made with AI watermark, cousin bottle in a new color.”

Review the pixels before they leave chat

Zoom to 400%. A commit diff hides a rewritten label. An overlay will not hide it either.

• Subject honesty. Same silhouette, color, label, and pack size. If a stranger would think this brand shipped a Copilot edition, discard the still.
• Job honesty. A README diagram stays in the repo. A UI mock is not a store screenshot. MAIN stays white, one unit, no graphics.
• Engine honesty. Write the model you used into the file name. A Copilot Flare still is Copilot Flare. A GPT Image 2 still is GPT Image 2. Do not stamp 2.5 on the FreeGPTImg JPEG.
• Empty type. Prices, percents, ratings, store chrome, and model names stay empty of anything you need people to read.
• Watermark honesty. If you opted into “Made with AI,” that file is not MAIN and not a listing screenshot. Turn it off for the next catalog pass.
• People and file. No recognizable generated faces you cannot clear. No children. PNG or JPEG, RGB, quiet field, no QR. Preview at feed size and at the size the README actually renders.
• Provenance. A Copilot public preview is not a license, not a marketplace approval, and not a signed offer. Recheck Amazon’s product image help and Meta’s ads guide before you tell a teammate the file is ready.

Limitations you should not skip

A Copilot still is not a signed discount, not a commercial-rights grant, and not MAIN. FreeGPTImg does not run GitHub Copilot, ChatGPT Images 2.5 Flare, or Sunburst. The default loop here is GPT Image 2 stills. Do not generate medical seals, “#1 2026,” store logos, or sale badges unless those claims are already on the real label and a manager signed them.

OpenAI’s image guide is explicit about the two jobs: generate from a prompt, or edit an existing image. Edits can be partial or entire. Treat that as a keep-subject loop, not as a promise Flare will honor an unsigned price. Quality tiers, background modes, and model IDs move. Recheck the live docs. ChatGPT Images 2.5, launched 8 September 2026, is stronger at keeping a subject from a reference photo. That is still not FreeGPTImg’s on-screen label.

Models invent letters, digits, and UI chrome when you mention an app store, a dollar figure, or “make it look shipped.” Shorten the brief. One SKU. One crop. Empty type. A new paragraph in Copilot is not a backup of the PNG you already had. If the locked raster is honest, recrop that. If it is not honest, rephotograph. Do not ask a chat still to remember a pack you never locked.

FAQ

What did GitHub ship on 28 October 2026?

Copilot can generate and edit images from natural language. The selected reasoning model calls a dedicated image tool powered by GPT Image 2.5 Flare, previews the result inline, and saves it to the project. It is in public preview for Copilot Pro, Copilot Pro+, Copilot Business, and Copilot Enterprise, billed on image tokens. Recheck the live changelog.

Does FreeGPTImg run GPT Image 2.5 Flare?

No. FreeGPTImg is a GPT Image 2 workspace: 20 daily free credits, no credit card required. Use it to lock the still. Write the name on the screen into the filename.

Can I use a Copilot image as Amazon MAIN or an App Store screenshot?

No. MAIN is a realistic photo of the item you sell, usually on pure white, without extra graphics or type. Store screenshots are captures of the real build, not a chat still with invented chrome. Keep Copilot output in the repo as a concept.

Should I turn on the “Made with AI” watermark for product stills?

No, not if the file might ship as MAIN, a listing, or an ad. GitHub’s watermark is opt-in and off by default. If a platform requires AI disclosure, follow that policy in copy or metadata. Do not bake a badge into pixels unless you meant the badge to be the product.

Why did Copilot invent a cousin kettle or fake 4.9 stars?

The prompt mentioned a listing, a price, “make it look live,” or the app store as scene dressing. Shorten it. Upload the locked photo on GPT Image 2. Name one crop. Empty type. Generate again.

Where should I lock catalog stills if Copilot already saved a JPEG?

Outside the chat. Open the photograph that matches the box. Run the keep / change / empty-type loop on GPT Image 2. Overlay signed words in an editor. Leave the Copilot file in docs/ with a concept filename.

Sources

• GitHub Changelog: Generate and edit images directly in Copilot
• OpenAI: Introducing ChatGPT Images 2.5
• OpenAI: Image generation
• Amazon: Product image requirements
• Meta: Image ads guide

Lock the still outside the chat

Pick the PNG that already matches the item you sell, not the JPEG Copilot dropped into the repo. Write five lines: keep, one crop or field change, empty type, overlay later. Open FreeGPTImg, stay on GPT Image 2 when that is the label on the screen, keep layout-only mocks in the AI UI generator, recrop a feed frame in AI social media ads, keep materials honest in AI product photography, cut a catalog crop in the Amazon product photo generator, run the edit in image to image, start a net-new concept in text to image only when there is no photo, and only then overlay the signed words.