How to Redo Your Photo in GPT Image 2 With One Word, No Secret Commands

A slash before a word in a GPT Image 2 prompt turns nothing on. It's just list formatting in other people's posts; the model reads the word as ordinary prompt text, not as a command to be recognized separately.

So "secret ChatGPT commands" don't exist. What exist are phrasings the model understands and takes into account during generation when you add them to your photo. The difference matters: you can wait years for a feature and never get it, but a phrasing you can copy and use right now.

Below are 8 phrasings for working with your own photo, tested on real shots, not abstract examples.

A series of shots from one photo

The phrasing photoseries turns one uploaded photo into a set of different angles and scenes based on it. Instead of generating one picture five times in a row, the prompt asks for a series at once, and the style, light, and composition stay the same across all frames.

It's handy for post covers or a carousel: one photoshoot becomes several different variants without reshooting.

Changing outfits from a photo of a look you like

You upload two photos, your own and the look you like, and ask the model to dress you in that look while keeping your face, pose, and proportions. The result doesn't paste someone else's face over yours; it transfers only the elements of clothing and style.

Every such prompt needs one more line: what must not change, meaning the face, the pose, and the composition. Without that line the model sometimes redraws more than you asked, and the result drifts further from the original photo than you wanted.

Changing the pose from a reference: pose and pose2

The phrasing pose transfers the pose from a reference photo to your shot; the model keeps the face and clothing and changes only the body position. If the first result doesn't suit you, the second phrasing, pose2, gives an alternative pose without rebuilding the prompt from scratch, with the same reference.

pose2 doesn't rewrite the whole prompt. It just offers a second variant, which saves time on repeat attempts.

Bird's-eye view: droneview

The phrasing droneview turns an ordinary shot into a view taken as if from a drone, from above. It works not only with portraits but also with product photos and landscapes in the background.

X-ray view: xray

The phrasing xray shows the inner structure of an object or scene instead of its outside. It's an unusual trick, it stands out in the feed, and it works as a visual hook for a carousel or a post cover about technology.

Glass effect: glassart

The phrasing glassart gives an artistic transparent effect and turns any object in the frame entirely into glass, with highlights and shimmer inside. It suits product shots and unusual covers where you need an effect that ordinary photo editing can't achieve.

Ad mockup: billboard and addcreative

The phrasings billboard and addcreative turn a personal photo into an ad mockup: the frame looks placed on a billboard or built into a ready ad creative. Useful for showing your product or service without a real studio shoot.

Cartoon style: cartoon plus a name

The phrasing cartoon with a specific style added, for example cartoondisney, converts the photo into the cartoon look of a recognizable studio. The style name is simply added after the word cartoon; nothing else needs to change.

Why a result sometimes gives itself away as AI

A generated face has three predictable tells. The skin is smoothed to a gloss with no pores or texture: fix it with phrasings from the film category, for example 35mmfilm or filmgrain, plus a direct phrase about natural skin texture without a plastic effect. The light is flat, with no direction and no shadow: softlighting, goldenhour, and rimlight help. And the face is too perfectly symmetrical, whereas living faces are almost always slightly off, which can also be written into the prompt as a separate phrase.

Bottom line

All eight phrasings above work not as a separate feature of the service but as ordinary words inside a text prompt to GPT Image 2. There's no hidden command behind them, only a precise phrasing the model understands better than a vague description.

I collected 106 such phrasings and 9 prompts for restoring dark, blurry, and old photos into one file, so you don't have to look them up again each time. Message СТИЛЬ to the Telegram bot, and I'll send the full list. More breakdowns on neural networks and images in the "Growing with AI" channel.