GPT Image 2.5 Prompt Generator

Plays to 2.5’s biggest gain — natural lighting and real material texture.

An editorial product photograph of [PRODUCT] on [SURFACE]. Lit by [LIGHT SOURCE, e.g. a single north-facing window] from camera [LEFT/RIGHT], soft directional falloff, roughly [COLOUR TEMPERATURE, e.g. 5200K]. The [MATERIAL, e.g. unglazed stoneware] reads as [SURFACE BEHAVIOUR, e.g. chalky matte with a faint speckle]. [FRAMING] at [LENS, e.g. 85mm] with shallow depth of field. Background: [BACKGROUND], softly defocused. Real photograph, honest texture, no retouching gloss, no watermark.

Keep word count low and hierarchy explicit — run this at xHigh or Max.

A vector infographic explaining [TOPIC]. Title "[TITLE]" across the top in bold [FONT STYLE]. Three numbered sections down the page, each a heading and one short line: "1. [HEADING] — [BODY]", "2. [HEADING] — [BODY]", "3. [HEADING] — [BODY]". Simple [ICON STYLE] icons beside each. [BACKGROUND COLOUR] background with [ACCENT] accents, thin rule lines, generous margins. Flat even lighting, every character sharp and correctly spelled, no duplicate text.

Matches the Poster template in ChatGPT Images 2.5.

A [STYLE] poster for [EVENT/BRAND], [SIZE/RATIO] format. Headline "[HEADLINE]" fills the upper third in [FONT STYLE], [COLOUR]. Subhead "[SUBHEAD]" beneath at roughly half the size. [IMAGE ELEMENT] occupies the middle. Detail line "[DATE / VENUE / URL]" along the bottom edge. Palette limited to [COLOURS], dominant colour [PRIMARY]. Clear hierarchy, generous margins, print-ready, no extra copy, no watermark.

Specify light source and skin texture rather than “beautiful lighting”.

A photorealistic portrait of [PERSON DESCRIPTION] wearing [WARDROBE]. [FRAMING, e.g. head and shoulders], body angled slightly camera left, eyes to lens, [EXPRESSION]. Lit by [LIGHTING SETUP] from [DIRECTION], [COLOUR TEMPERATURE]. Natural skin texture with visible pores, fine flyaway hair, no beauty retouching. Background: [BACKGROUND], softly defocused. 85mm feel, sharp focus on the near eye, true-to-life colour.

Subject preservation from reference photos is a headline 2.5 upgrade.

Image 1: the base scene — preserve its composition, lighting direction and colour grade exactly. Image 2: the subject — [WHO/WHAT], this is the identity source. Place the subject from Image 2 into Image 1 at [POSITION], scaled so [SCALE CUE, e.g. their head reaches the window sill]. Match Image 1’s lighting direction, colour temperature and grain. Preserve: face, identity, outfit, proportions. Constraints: no extra objects, no added text, no halos or fringing.

2.5 changes only what you ask — name the change and the preserve list.

Change: [SPECIFIC ELEMENT] becomes [NEW STATE]. Preserve: composition, framing, lighting, shadows, colour grade, every other object, and all existing text exactly as they are. Constraints: no redesign, no extra objects, no added text, no watermark. Blend the edited area so its lighting, grain and focus falloff match the rest of the frame.

Quality holds across turns — make one change per turn instead of one compound edit.

Turn 1 — Change only [FIRST CHANGE]; everything else stays exactly the same.
Turn 2 — Now change only [SECOND CHANGE]; same person, same outfit, same framing as the previous image.
Turn 3 — Now change only [THIRD CHANGE]; keep the colour grade and lighting from the previous image.
On every turn preserve: [FACE / POSE / LAYOUT / BRAND TEXT]. Never add objects or text that were not requested.

For the images/edits endpoint with an alpha-channel mask.

Fill the masked region only with [NEW CONTENT]. Match the surrounding [GRAIN / NOISE LEVEL], focus falloff, lighting direction and colour temperature so the seam is invisible. Do not alter a single pixel outside the mask. Constraints: no halos or fringing at the mask edge, no added text, no extra objects.

Transparency improved in 2.5 — requires PNG or WebP output.

A [PRODUCT/SUBJECT] rendered as a clean cutout, [ANGLE, e.g. three-quarter view], lit by [LIGHTING] so the form reads clearly without a backdrop. Material: [MATERIAL] with [SURFACE FINISH]. Background: fully transparent (alpha channel), subject cleanly cut out, crisp edges on [FINE DETAIL, e.g. hair / wire / glass rim], no halos, no fringing, no drop shadow baked in.

Matches the Merch template; keep it flat and print-ready.

A [STYLE] graphic for [GARMENT], centred composition. Main motif: [MOTIF]. Lockup text reads "[TEXT]" in [FONT STYLE], [COLOUR], positioned [PLACEMENT]. Limited to [NUMBER] flat colours: [COLOURS]. Consistent stroke weight, no gradients, no drop shadows, clean vector edges, transparent background, print-ready at large scale.

For the @Sketch tool — say what the drawing governs and what it does not.

Use the supplied sketch for layout and proportion only — the final image is a [TARGET MEDIUM, e.g. photorealistic interior photograph], not a drawing. Scene: [SCENE]. Subject: [SUBJECT] positioned as sketched. Render in [STYLE] with [LIGHTING] and [MATERIALS]. Keep the sketched spatial relationships and scale; ignore the sketch’s line quality, colour and shading.

Preserve typographic hierarchy across scripts; keep total word count low.

A [SIGN TYPE] displaying "[TEXT IN TARGET LANGUAGE]" in [SCRIPT, e.g. Japanese kanji and kana] set in [FONT STYLE], with "[SECONDARY TEXT]" beneath in smaller English at roughly half the size. Mounted on [SURFACE] in [SETTING]. [LIGHTING]. Straight-on framing, high contrast, text rendered crisply and correctly in both scripts, no duplicate or garbled characters.

Lock one specification, then reuse it verbatim across every panel or page.

Character anchor — [NAME]: [AGE/BUILD], [HAIR], [SKIN], wearing [OUTFIT DETAIL], [DISTINGUISHING FEATURE]. Proportions: [PROPORTION NOTE, e.g. three-and-a-half heads tall]. Style: [ART STYLE], [LINE WEIGHT], [PALETTE].
Scene for this panel: [SCENE AND ACTION]. Preserve identity exactly: same face, same outfit, same proportions and same palette as the anchor above. No style drift between panels.

Consistent multi-element output in one generation.

A set of [NUMBER] [ASSET TYPE, e.g. app icons] for [BRAND], arranged in an evenly spaced grid. Each depicts: [ITEM 1], [ITEM 2], [ITEM 3], [ITEM 4]. Consistent stroke weight, corner radius and optical size across all of them. Palette limited to [COLOURS], dominant colour [PRIMARY]. Plain [BACKGROUND] background, flat even lighting, crisp edges, no text labels.