The GIF Is a Loop. Don't Ship It as the SKU.

OpenAI’s image guide now documents animated GIF output on GPT Image 2.6. The family has two Image API models: gpt-image-2.6-sunburst for detailed visual work and gpt-image-2.6-flare for fast iteration. Both can return a GIF. Set output_format: "gif" on POST /v1/images/generations or POST /v1/images/edits. Describe the subject, the motion, and the loop in the prompt. To guide that loop with still photographs, send the photographs to the edit endpoint. frame_count is 2–16 and defaults to 9. frame_duration_ms is 20–2000, rounded to 10 ms, and defaults to 200. OpenAI’s own lantern example uses nine frames at 200 ms; the monkey returns to its starting pose after 1.8 seconds. That file is a loop. It is not Amazon MAIN. It is not a Meta feed still. Catalog and campaign stills still start 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 2.6, Flare, Sunburst, or .gif on a GPT Image 2 JPEG, and do not export the lantern as the product.

Draft in FreeGPTImg. Stay on GPT Image 2 when that is the label on the screen. Recrop a feed frame in AI social media ads. Keep materials honest in AI product photography. Put one unit on white in the Amazon product photo generator. Change only the field in white background product photo. 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

OpenAI’s image guide still splits the work: generate from a prompt, or edit an existing image, including partial edits. GPT Image 2.6 sits on that Image API path. The model pages list /v1/images/generations and /v1/images/edits as supported. They list Responses as not supported. Do not send a 2.6 ID into a Responses thread and expect the same handles as 2.5. The same guide still says the examples below the 2.6 section use GPT Image 2.5. Write the ID that actually ran.

Both 2.6 models support complexity control, markup-guided edits, SVG output, and animated GIF output through the Image API. Complexity is how much the model plans. Markup is a circle on a transparent PNG. SVG is for icons and illustrations with editable paths. GIF is a loop you describe, then save as .gif. None of those formats is Amazon MAIN. None of them is a Meta feed still you overlay later.

GIF is the unused handle. Recheck the live guide if you still think GIFs are generations-only. They are not. Generations invent a loop from a prompt. Edits can take still reference images and animate from those pixels. That is the only honest path when the object already exists in a warehouse. frame_count and frame_duration_ms work only with GIF output. GIF does not support streaming, partial images, custom output_compression, or prompt_rewriter. size describes each finished frame: auto, or WIDTHxHEIGHT with positive multiples of 16. If the request blows the pixel budget, reduce frame_count, the frame size, or n. Nine frames at 200 ms is 1.8 seconds. Sixteen frames at 2000 ms is a different file. Neither file is the carton.

Amazon’s product image help still treats MAIN as a realistic photo of the item you sell, usually on a pure white field, without extra graphics or type. A looping lantern is not MAIN. Meta’s image-ad guide describes feed image ads as still image files, not a nine-frame blink. Recheck both pages before you tell a teammate the GIF is ready to upload.

The cheap failure today is not “I did not know 2.6 can loop.” It is “I asked for a GIF of the kettle, the spout rewrote itself on frame four, and I shipped the loop as the pack.”

Name the still before you generate

Weak sets fail because one 2.6 request tries to be a concept still, a GIF, MAIN, a 4:5 ad, an SVG mark, and the signed headline. Pick one job.

Job	What you export	Keep	Change one thing	Wrong move
Concept still	A mood board or placeholder	One idea	Nothing else	Feeding it back as the product
Locked catalog still	The SKU you already ship	Shape, color, label, logo, pack size	Nothing, or light only	Starting from a generated JPEG
White MAIN draft	Same SKU, square, opaque white	The upload	Field only	Chat chrome, extra unit, a GIF
4:5 feed still	Same SKU, empty lower or top band	Silhouette, palette	Crop and empty band	Asking the model to “make it move”
9:16 story still	Same SKU, empty lower third	The object	Crop only	Status bars, prices, “GIF” badges
Focused raster edit	Same locked photo	Everything the prompt does not name	One noun	Turning the edit into a loop
GIF loop	A .gif, 2–16 frames	Motion you described, or the locked photo if you sent it	Nothing that must match a listing	Using the lantern as the pack shot
SVG mark	A vector icon or wordmark	Paths, not a photo	Nothing about MAIN	Animating the SVG as a product shot
Offer overlay	Don’t generate it	Empty type	Editor, last	“SALE”, “$29”, “Shop now” in the prompt

A text-to-image GIF invents. An image-to-image GIF should borrow. If you already paid for a pack shot that matches the listing, that file is truth. The 2.6 loop is extra motion. It is not a second product.

Checklist before anything leaves the thread

Use this sequence. Do not paste a campaign brief onto output_format: "gif" and expect the export to be the SKU you used to ship.

1. Name the job. Concept still, catalog still, MAIN draft, feed still, story still, focused raster edit, GIF loop, SVG, or overlay. One file, one job.
2. If it is a catalog, ad, or listing 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 the prompt asked the lantern to blink.
3. Write keep / change / do-not in five lines. Keep silhouette, color, label, logo, and pack size. Change only one noun: crop, field, empty band, or — if this file is actually a GIF — one motion. Do not add a second SKU, a restyled colorway, or a percent.
4. Name the engine on the screen. If FreeGPTImg shows GPT Image 2, write gpt-image-2. If an API log says gpt-image-2.6-sunburst or gpt-image-2.6-flare, write that ID. If the export is a GIF, write gif, frame_count, and frame_duration_ms into the filename too. Do not relabel either file 2.6-approved.
5. Lock the still outside 2.6 first. Run the keep / change / empty-type loop on GPT Image 2. Only then send that PNG into a 2.6 GIF, and only through the edit endpoint if the object must stay the SKU.
6. Do not text-to-GIF a shipping still. Generations invent. Edits can take still references. OpenAI is explicit: to guide the animation with still images, send them to /v1/images/edits. If the photograph is already truth, that is the input. If you do not yet have a photo, keep the GIF in the concept folder.
7. Ask for empty type. Prices, percents, star ratings, “Shop now,” 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. A looping “SALE” is still a claim.
8. Describe the loop, not a new object. Subject, motion, return to the start. OpenAI’s lantern returns to its starting pose. If frame nine is a cousin lid, the loop failed.
9. Pin the GIF fields. Set output_format: "gif". Set frame_count in 2–16. Set frame_duration_ms in 20–2000. Do not omit them and then call the default nine-frame 200 ms file “final.” If the request exceeds size limits, shrink frames, size, or n — do not raise quality and hope.
10. Inspect every frame before the overlay. A rewritten mark on frame three, a floating “GIF”, a ribbon that says “SALE”, or a tag that says “4.9” are stop conditions. Hold the first and last frame next to the real pack. If a stranger would think this brand shipped a blinking edition, discard it.
11. Only then overlay, and only on a still. In an editor, set the real headline on a locked raster, not inside the GIF pixels. Recrop to the live placement size after the still matches. Do not ask Sunburst to spell the price across nine frames. Do not ask Flare to invent a lid that “reads cleaner in motion.”

Prompt formula for the locked still

GPT Image 2 catalog or ad still, not a 2.6 GIF 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 / one noun].
Empty type. No prices, percents, ratings, model names, watermarks, or chat chrome.
Do not add: extra SKUs, restyled colorways, “Shop now”, celebrity faces, children.

Prompt formula for the GIF after the still is locked

Use this product photograph exactly as uploaded. Do not restyle, recolor, relight, or redraw the SKU.
Keep labels, logo, silhouette, and pack size accurate in every frame.
Motion only: [one loop, return to the start].
Leave all type empty. Do not invent prices, percents, or CTAs.
output_format: gif
frame_count: 9
frame_duration_ms: 200

Prompt examples

Locked catalog still. “Image-to-image: keep this exact matte-black kettle, spout curve, handle, and printed mark. Even studio light. Empty type. Not a 2.6 GIF, not MAIN unless named.”

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

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

9:16 story, empty lower third. “Image-to-image: keep this exact mug glaze and printed mark. 9:16, empty lower third. No status bar, no ‘Shop now’, no model watermark.”

GIF from the locked photo. “Image edit: use this product photograph exactly as uploaded. Nine frames, 200 ms, kettle steam only, return to the start. Do not restyle the spout, mark, or lid. Empty type.”

Do not generate this. “GIF of a new copper kettle, ten colorways, CHAMPIONS banner, Instagram CTA Shop now $29, 4.9 stars, guaranteed Amazon approved, cousin lid that looks cleaner in motion, SVG of the bottle, MAIN white background.”

That last prompt is four jobs and a lie. Split it.

Review the pixels before they leave the overlay

Zoom to 400%. A loop hides a rewritten label on the middle frames. An overlay will not hide it either.

• Subject honesty. Same silhouette, color, label, and pack size on frame 1 and frame n. If a stranger would think this brand shipped a 2.6 edition, discard the file.
• Job honesty. A concept still stays in the concept folder. A GIF is a GIF until you prove otherwise. MAIN stays white, one unit, no graphics, not animated. An SVG stays a mark.
• Engine honesty. Write the model you used into the file name. A GPT Image 2 still is GPT Image 2. A 2.6 Sunburst GIF still is Sunburst plus output_format:gif. Do not stamp 2.6 on the FreeGPTImg JPEG.
• Empty type. Prices, percents, ratings, store chrome, and model names stay empty of anything you need people to read. Motion is not a headline.
• Handle honesty. frame_count 2–16. frame_duration_ms 20–2000. Generations invent. Edits take stills. GIF output does not stream. If you needed a new object, you did not need a product GIF.
• People and file. No recognizable generated faces you cannot clear. No children. Preview the loop at the size it will actually play, and preview the locked still at the size the catalog actually renders.
• Provenance. A 200 from the Image API 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 GIF is not a signed discount, not a commercial-rights grant, and not MAIN. FreeGPTImg does not run GPT Image 2.6, Flare, Sunburst, or GIF output. 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 own limitations still apply: complex prompts can take up to two minutes; text in the image can still miss placement and clarity; recurring brand marks can drift across generations; structured layouts are harder than they look. Those limits get worse across frames. Treat them as reasons to lock the photograph first, not as a dare to raise frame_count.

The 2.6 model pages currently support image generation and image edit. They do not list Responses. GIF needs output_format: "gif" on generations or edits; frame_count is 2–16 (default 9) and frame_duration_ms is 20–2000 (default 200). SVG is a different export. Complexity is a different field. Neither format is a listing photo. Recheck the live docs; model IDs, frame limits, and output formats move.

Models invent letters, digits, and product cousins when you mention Instagram, a dollar figure, “make it move,” “more premium,” or “make ten.” Shorten the brief. One SKU. One motion. Empty type. A nine-frame 200 is extra motion, 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 lantern to remember a pack you never locked.

FAQ

What is GIF output on GPT Image 2.6?

It is an animated loop from the Image API. Set output_format: "gif" on generations or edits. Describe the subject, motion, and loop. frame_count is 2–16 (default 9). frame_duration_ms is 20–2000, rounded to 10 ms (default 200). To guide the loop with still photographs, send them to the edit endpoint. Recheck the live image-generation guide.

Can I use a 2.6 GIF as Amazon MAIN or a Meta image ad?

No. MAIN is a realistic photo of the item you sell, usually on pure white, without extra graphics or type. A restyled glaze, a blinking lid, or baked “Shop now” type fails that job. Feed image ads are still-image files. Lock the SKU first. Empty type. Overlay last. Recheck Amazon’s product image help and Meta’s ads guide.

Does FreeGPTImg run GPT Image 2.6, Flare, Sunburst, or GIF output?

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.

Should I generate the GIF from a prompt or from a locked photo?

From a locked photo if the object must stay the SKU. OpenAI’s guide says still reference images go to the edit endpoint. Text-to-GIF is a concept. Image-edit GIF is a loop of pixels you already had. Keep the concept file in the concept folder.

Why did the spout change on frame four?

Because a loop is extra motion, not a lock. Brand marks can drift across generations. If the photograph is already truth, send that PNG, name one motion, and tell the tool not to restyle the rest of the SKU. If frame nine is a cousin lid, discard the GIF.

Should I lock stills in 2.6 or outside it?

Outside it, if the file might ship as catalog, MAIN, or an ad. Photograph the item. Run the keep / change / empty-type loop on GPT Image 2. Then, if you still need a GIF, send that PNG to a 2.6 Image API edit with output_format: "gif" and tell the tool not to touch the rest of the SKU. Do not send 2.6 into Responses; the current model pages do not list that endpoint.

Sources

• OpenAI: Image generation
• OpenAI: GPT Image 2.6 Sunburst
• OpenAI: GPT Image 2.6 Flare
• Amazon: Product image requirements
• Meta: Image ads guide

Lock the still before you export the loop

Pick the PNG that already matches the item you sell, not the GIF a 2.6 pass invented because the prompt asked a lantern to blink. Write five lines: keep, one motion, empty type, overlay later. Open FreeGPTImg, stay on GPT Image 2 when that is the label on the screen, recrop a feed frame in AI social media ads, keep materials honest in AI product photography, put one unit on white in the Amazon product photo generator, change only the field in white background product photo, 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 — or send the locked PNG into a 2.6 Image API edit with output_format: "gif", never a text-to-GIF cousin as the pack, and tell the tool not to touch the SKU.