How-to-Use Product Photos: One Real Action, Empty Step Bands

A how-to-use product photo is a secondary listing still of one real action on the SKU you sell: unfold, seat, fill, press, clip. It is not MAIN, not a lifestyle room, and not a “3 easy steps” poster with invented captions. Lock the product, the true motion, and the number of frames first. Leave step numbers and instruction lines empty for an editor. Amazon’s product image help (as of 2026) still wants MAIN to be a realistic image of the actual product on a white field (RGB 255, 255, 255), the item filling most of the frame, JPEG/PNG/TIFF, and enough pixels for zoom (plan 1,000 px or more on the long edge). Additional images can show the product in use and infographic layouts. OpenAI launched ChatGPT Images 2.5 on September 8, 2026, and said it keeps a subject more stably while you change one element—use that to move a lid, not to invent a second gadget. Amazon set Prime Big Deal Days for October 6–7, 2026; setup stills need to match the carton before that window, not wear fake “easy 3-step” bursts. FreeGPTImg is a GPT Image 2 workspace: 20 daily free credits, no credit card required. Overlay real instructions last.

Draft the still in FreeGPTImg. Keep materials honest in AI product photography, cut a listing crop in the Amazon product photo generator, lock a clean field in the white-background product photo tool, upload a locked hero into image to image when you only change the pose, stage one in-use crop with the lifestyle product photo generator, stay on GPT Image 2 for the loop in front of you, invent the first layout in AI text-to-image, and leave number boxes empty in the AI infographic maker. Do not put a three-panel board on MAIN.

What a how-to-use still actually does

Shoppers opening a travel kettle, a phone clamp, or a French press are answering “can I actually run this?” The still answers with the same product in one honest motion. MAIN answers with the product as sold—usually assembled, on white, filling the frame. A three-panel board answers the same motion split into empty-numbered frames. A lifestyle crop answers one in-use moment in a real room. A cartoon hand, a second SKU “for context,” and a baked-in “STEP 1 BOIL 3 MIN” line answer a manual you did not write.

The motion is a SKU fact, not a mood. If the dripper unfolds, do not generate a motorized base. If the clamp ships without a Bluetooth remote, do not invent one because “setups usually include it.” If the press has three real actions—add, pour, press—do not pad a fourth “enjoy” frame with a croissant the carton does not contain. A what's-in-the-box still is a different slot: parts that ship, counted. A how-to still is the product being used, not a packing list.

ChatGPT Images 2.5 can keep a subject while you change one object. That is the sequence job: keep the kettle, open the lid. It is not permission to add a copyrighted mascot, a competitor’s travel pouch, or a Prime smile. FreeGPTImg runs GPT Image 2. Write the model you used into the file name.

Do not prompt studio wordmarks, celebrity hands demonstrating a trick, or “as seen on” badges. Photorealistic AI people are a separate policy: if a marketplace requires a contains-synthetic-performer tag in IPTC/XMP, apply it in an editor. Prefer a tabletop, a packing-paper field, or unrecognizable hands.

Pick one slot before you generate

Weak how-to sets fail because they ask for a white MAIN, three numbered steps, a lifestyle kitchen, and a sale burst in one pass. Name the slot first.

Slot job	Working crop	Start from	Change one thing	Keep	Wrong move
Single in-use crop	Square or 4:5	Locked hero of the same SKU	Pose / grip only	True silhouette, empty caption	Extra gadgets, fake steam gauges
3-panel how-to board	4:5 or 16:9, empty numbers	Same SKU, same lighting	Layout into panels	Empty numerals, empty verbs	Model-written “Step 1,” times, temps
Assembly still	Square	Only parts that ship	Arrangement only	True count, empty names	Bonus tools you do not pack
Safety / warning band	4:5, empty type	Same product	Empty band only	Empty warning copy	Invented “non-toxic,” CE marks
MAIN (not how-to)	Square, white field	Product as sold	Crop only	Pure white, no panels	Putting the board on MAIN
Social recrop	1:1 or 4:5 or 9:16	One how-to still	Crop only	Empty headline	Baking “3 easy steps” into pixels

MAIN is not a how-to board. A three-panel still is not MAIN. A lifestyle still is not a manual. Generate large, then recrop in an editor to the box the channel shows. Overlay real words later. Step verbs, times, temperatures, URLs, prices, “limited-time,” and “bonus” never belong in the generator if they must be exact.

How-to-use workflow

Use the same sequence in product-photo tools, an image-to-image pass, a GPT Image 2 tab, or ChatGPT Images 2.5.

1. Write the motion in five lines. SKU truth (one offer, true materials, true pack count), the real action in order, slot name, crop, empty type zones. If a step is not how the product works, it is not in the prompt.
2. Lock one as-sold hero. White field, even light, product filling most of the frame. This still is MAIN. Do not put panels here.
3. Write keep / change / do-not. Keep silhouette, materials, finish, and the exact SKU. Change only the pose or the panel layout. Do not add logos, seals, QR codes, Prime marks, ranking bursts, extra products, or extra accessories.
4. Generate one in-use crop from the locked hero. One pass, one action. File names like kettle-howto-pour.jpg. If turn two adds a mug, revert to the locked hero and try again.
5. Optional three-panel board. Same lighting, same SKU, large empty number rectangles. Do not let the model write “1 unfold,” “2 fill,” or “3 press.” Overlay those in an editor after a merchandiser checks the real manual.
6. Match the crop. Square for many gallery files. 4:5 for a tall board. 16:9 only if that is the module you actually upload. 1:1 or 9:16 for a social recrop. Do not stretch a white MAIN into a Story and call it a how-to.
7. Inspect the motion before mood. A Bluetooth remote you do not ship, a second bottle, a croissant “for vibe,” or a timer reading 3:00 are stop conditions. Hold the still next to the real product. If a stranger could not perform that action with what is in the carton, discard it.
8. Overlay copy last. In an editor, set the real step verbs and only the claims a manager signed. Keep marketplace files in the format that channel lists (often JPEG or PNG, RGB, enough pixels for zoom).

OpenAI’s Images 2.5 notes still help if you never open ChatGPT: keep the subject, edit one element, expect earlier choices to survive later turns. That is a sequence rule. Open the lid. Do not invent a second kettle.

Amazon’s October 6–7, 2026 Prime Big Deal Days window is a deadline for honest setup stills, not a brief to stamp “Lightning Deal” or “3-step miracle” on the pixels. Deal copy lives in Seller Central. The photo still has to look like the product that ships.

Prompt formula

Catalog how-to-use still, not a MAIN hero and not a “3 easy steps” poster.
Product: one [SKU], accurate silhouette, materials, scale, and pack count.
Motion: [the real action in order], true to the manual, no extra gadgets “for context.”
Slot: [single in-use / 3-panel board / assembly / safety band / MAIN as-sold / social recrop].
Light: even catalog light for boards; one honest indoor lamp or window for a single in-use crop.
Keep: empty step numbers, empty verb bands, empty time/temp boxes.
Change only: [the pose / the panel layout].
Text in image: none. No step captions, times, temperatures, prices, QR codes, watermarks, or marketplace badges.
Do not add: extra SKUs, extra accessories, celebrity faces, Prime logos, sale bursts, food props not in the carton.

Prompt examples

Single in-use, travel kettle. “Square catalog still of one stainless 600 ml travel kettle, lid open, even studio light, quiet white or light-gray field. One unrecognizable hand pouring into the spout. No mug, no bag of beans, no travel pouch, no readable slogans, no Prime badge, no ‘step 1’ type.”

Three-panel board, empty numbers. “4:5 catalog board of the same kettle, identical lighting, quiet white field. Three equal panels: lid closed, lid open, pouring. Large empty number rectangles in each panel. No printed verbs, no ‘3 minutes,’ no prices, no ‘new’ bursts.”

Image-to-image, pose only. “Image-to-image: keep this exact kettle silhouette, steel, and studio lighting. Change only the lid from closed to open. Quiet white field. No extra objects, no text, no second kettle.”

MAIN, as sold. “Square catalog still of the same kettle standing assembled, lid closed, true steel, even studio light, pure white background RGB 255 255 255. Product fills most of the frame. No panels, no pouring, no readable slogans.”

Social recrop. “9:16 recrop of the single in-use still, kettle centered in the middle third, empty lower band for later overlay. No ‘3 easy steps’ type, no percentages, no platform UI chrome.”

Review the pixels before you upload

Zoom to 100%. A phone grid hides a second bottle.

• Subject honesty. Same silhouette, lid, fabric, and SKU. Only the motion you named. If a stranger would think this is another brand’s deluxe kit, discard the still.
• Slot honesty. White MAIN stays as-sold and unpaneled. How-to stays off MAIN. Lifestyle shows one hero item unless the SKU is a set you actually pack.
• Empty type. Step verbs, times, temperatures, “3 easy steps,” prices, and marketplace badges stay empty of anything you need people to read.
• People and file. No recognizable generated faces you cannot clear. JPG or PNG, RGB, enough pixels for zoom, no watermark, no QR. Preview on a phone and as a square.
• Provenance. OpenAI still uses C2PA metadata and an invisible watermark on GPT Image outputs. Downstream apps may keep, strip, or flag that data. A prettier pour is not a license.

Limitations you should not skip

A generated still is not marketplace approval, not a user-manual substitute, and not a commercial-rights grant. FreeGPTImg does not upload into Amazon, Shopify, or a retailer CMS. Do not upload someone else’s instruction photo “to see the style,” and do not imitate a competitor’s remote. If a platform requires AI disclosure, follow that policy. ChatGPT Images 2.5 is not GPT Image 2; on FreeGPTImg you are in a GPT Image 2 loop. If turn three adds a mug, revert. Do not generate medical, “non-toxic,” boil-time, or “best of 2026” badges unless those claims are already on the real label and a manager signed them. Recheck live marketplace help for MAIN background, file type, additional-image rules, and pixel floors before you publish.

FAQ

What is a how-to-use still?

It is a secondary catalog still of one real action on the product you sell: a single in-use crop, an optional three-panel board with empty number bands, and a recrop if you need social. It is not MAIN. MAIN is the product as sold. FreeGPTImg runs this on GPT Image 2, with 20 daily free credits and no credit card required.

What size should I generate?

Do not treat a social crop as your only canvas. Plan 1,000 px or more on the long edge for marketplace files, square for many gallery stills, 4:5 for tall boards, and 1:1 or 9:16 only as recrops. Confirm the live help article for the channel you publish to. Stretching a white MAIN into a Story is how the action falls out of the frame.

Should I put “Step 1: boil 3 minutes” in the image?

No. Step verbs, times, and temperatures belong in the manual or overlay after a merchandiser checks the real instructions. Leave those rectangles empty. Invented times are a review risk and a returns problem.

Can the how-to board be MAIN?

Usually no. MAIN is the product as sold on a white field, without extra props or numbered panels. Put the sequence in an additional-image slot. If the offer is a set, MAIN can show the set as sold—not a comic of spare parts plus food you do not pack.

Can I add a bonus tool for the October sale?

No. Prime Big Deal Days on October 6–7, 2026 is not a reason to generate a remote, mug, or pouch you cannot ship. Generate only the product and the real motion. Deal copy stays out of the pixels.

Why did the model add a croissant next to my French press?

The prompt asked for a “complete breakfast setup” or “cozy kitchen.” Shorten it. One offer. One action. Empty type. Generate again. If extras still appear, lock a white-background hero of the real product first, then run image-to-image with “change pose only, add nothing.”

Sources

• Amazon Seller Central: Product image requirements
• About Amazon: Prime Big Deal Days is back October 6-7
• OpenAI: Introducing ChatGPT Images 2.5
• OpenAI: Image generation
• C2PA: Content Credentials

Make one how-to-use still on FreeGPTImg

Write five lines for one offer and the real motion. Ask for empty step boxes. Open FreeGPTImg, run the job in AI product photography or the Amazon product photo generator, keep MAIN white in the white-background product photo tool, change pose from a locked hero in image to image, leave number bands empty in the AI infographic maker, stay on GPT Image 2 when that is the model in front of you, and overlay the real instructions only after the still matches the product you ship.