GPT Image 2.5 prompt — a product billboard mockup, refined across two turns
Build the scene in turn one. Change one condition in turn two. Five words is enough for the second.
| Mode | Model | Quality | Size | Ratio | Credits |
|---|---|---|---|---|---|
| Image to Image | Think | Standard | 1K | 2:3 | 10 cr |
Published by OpenAI as an example for this model — not generated on this site. Model: gpt-image-2.5-sunburst. Source: OpenAI The published example ran at 1024x1536, a legal size we do not currently offer; the frame here is the same shape at our nearest size.
The shampoo bottle again, this time on a highway billboard at sunset carrying the line "Fresh and clean" — and then, in a second pass of five words, at night in snow. This is the template that teaches iteration: pass the output back in as the input, change exactly one thing, and keep going. The second prompt is short because everything else is already established.
The prompt, nothing cut
Create a realistic billboard mockup of the shampoo on a highway scene during sunset. Billboard text (EXACT, verbatim, no extra characters): "Fresh and clean" Typography: bold sans-serif, high contrast, centered, clean kerning. Ensure text appears once and is perfectly legible. No watermarks, no logos.
Turn two
Turn two — change one condition
Pass the billboard image back in as the input. Five words, and the scene, the product and the copy all survive because none of them were mentioned.
Make it look like a winter evening with snowfall.
This exact prompt, at these exact settings
The mode and all five settings are locked to this template, so there is nothing to line up by hand. Press Run this prompt above and the panel opens filled in — prompt, mode, quality, size and frame — and your first image needs no account. If your plan sits below what this ran at, the line under the panel says so.
This example was published at Think · Standard · 1K. A free run is Fast · Standard · 1K, so expect less fine detail and softer small text. The composition is the same; the finish is not.
What is safe to change
Most libraries publish this list and stop, which is why so many copied prompts come back worse than the original: the swappable nouns are the safe half. The three sections under it are the other half — the clauses that are doing the work, the ones that break it if you touch them, and why it is written in this order.
The text fence
"(EXACT, verbatim, no extra characters)" plus the quoted line plus "Ensure text appears once". Three overlapping guarantees, which is what billboard copy needs because the type is the largest thing in the frame.
The typography line
Weight, contrast, alignment, kerning — described as properties rather than as a font name. This is the reliable way to direct type on this model.
The follow-up length
Five words. The rule for turn two is to change one thing and mention nothing else, because anything you re-describe is something you have re-opened.
Three ways to break it
Each one is a real failure with a reason attached. A rule without a reason is not usable.
01Restating the whole scene in turn two
Counter-intuitive but reliable: re-describing the billboard, the product and the copy in the follow-up gives the model permission to reinterpret all three. Say only what changes.
02Changing two conditions at once
When a two-change turn comes back wrong you cannot tell which change caused it. OpenAI's guidance is to request one change and compare before adding more instructions.
03Assuming the copy survives every turn
It usually does, and it does not always. Repeated edits can move details you intended to preserve. Read the billboard text again after every turn — it is the thing you would ship and the thing you would stop looking at.
Inside the Put your product on a billboard, then change the weather prompt
The point of this template is not the billboard. It is the shape of the second prompt.
Turn one is a full brief: the scene, the product, the copy fenced three ways, the typography, the exclusions. Turn two is "Make it look like a winter evening with snowfall." Five words, and the billboard survives, the shampoo survives, the text survives, the composition survives — because none of them were mentioned. The previous image is the input, and everything in it is the default.
That is the whole discipline of iterative editing and it runs against instinct. When people want a change they tend to re-describe the target state in full, on the reasonable theory that more information is safer. It is not. Every element you re-describe is an element you have re-opened for interpretation. Mention the shampoo bottle in turn two and the model considers how to render a shampoo bottle; leave it out and it simply carries.
OpenAI's guidance says the same thing from the other direction: pass the previous output as the next edit input, request one change, and repeat the details to preserve if the result drifts. Note the conditional. Restating constraints is a repair for observed drift, not a precaution to apply by default.
The one-change-per-turn rule has a second justification beyond reliability, which is diagnosis. When a turn comes back wrong and it contained one change, you know what caused it. When it contained three, you are back to guessing, and the cheapest way out is to start over.
Turn one is also the strongest example of text fencing in the library. Billboard copy is the largest text in the frame and the first thing anyone reads, so it gets three overlapping guarantees: the parenthetical "(EXACT, verbatim, no extra characters)", the quoted string, and "Ensure text appears once and is perfectly legible". Redundant on paper. Not redundant in practice, and worth the tokens on anything where a typo means starting again.
Put your product on a billboard, then change the weather — common questions
- Why is the follow-up prompt so short?
- Because the previous image is the input and everything in it is the default. Anything you re-describe is something you have re-opened for reinterpretation.
- Can I make two changes in one turn?
- You can, but when it comes back wrong you will not know which change did it. One per turn, compare, then continue.
- Will the billboard text survive several turns?
- Usually. Not always. Repeated edits can move details you meant to keep, so re-read the copy after every turn.
- How many turns can I chain?
- As many as the result stays good. Quality degrades gradually rather than failing, so compare against the original every few turns rather than only against the previous one.
Other prompts that start the same way
- Think · Standard · 1K · 3:2
Input 
Result Swap one piece of furniture in a real room photo
A capitalised ONLY plus a list of photographic properties that must survive the swap.
Bring an image10 cr - Think · Standard · 1K · 2:3
Input 
Result Draw the next page without the character changing
A new scene with the character constraints restated verbatim from the reference prompt.
Bring an image10 cr - Think · HD · 1K · 2:3
Input 
Result Translate the text in an image without moving the layout
Fourteen words. Half of them exist to stop the model redrawing the design.
Bring an image28 cr
Written and maintained by Andy SwiftPublished Last updated