GPT Image 2.5 Tutorial

Make a skincare campaign image, change its background without redesigning the product, then adapt it for a vertical post. Follow the three prompts below in LumiYing and use each chosen result as the reference for the next step.

Try GPT Image 2.5 GPT Image 2 vs GPT Image 2.5

What is GPT Image 2.5?

GPT Image 2.5 creates images from text and edits existing images from a written instruction. Its improvements in reference fidelity and focused editing are especially useful when you want to keep a product or person recognizable while changing the scene around them.

This tutorial uses one fictional skincare product throughout. You will practice three different tasks: describe a complete image, change one part of it, and adapt its composition to another format. You can later apply the same approach to your own product photos.

Make one campaign image, then create two variations

Keep the same product and headline throughout the exercise.

Open LumiYing in image mode and choose GPT Image 2.5 (Flare). Set 1K, Medium, 1:1 and one image. Start without a reference for this fictional product. If you already have a real product photo, upload it and replace the bottle description in the first prompt with an instruction to preserve that product.

Step 1

Create the square campaign image

Describe the product, its placement and the headline as one complete composition. Here, the bottle sits in the lower half and the text occupies the top. That gives the model a specific layout to work with.

Create a square skincare campaign poster. Show one frosted amber glass serum bottle with a matte ivory dropper cap on a pale stone surface. The bottle label reads exactly "SERUM". Center the bottle in the lower half of the frame. Put the exact headline "DAILY CARE" at the top in large dark charcoal letters. Use a warm gray studio background and soft window light from the left. Keep generous space between the headline and bottle. No extra words, props or decorative graphics.

Look for a complete bottle, a readable label and clear space below the headline. If the layout is crowded, fix that before moving on. Increasing the resolution will not give the headline more room.

Step 2

Change the background, keep the product

Save the version you like and upload it as the reference. Keep 1K, Medium and 1:1. Ask for the background change explicitly and name the details that should remain unchanged; a request such as “make it fresher” leaves too much open.

Edit the uploaded poster. Change only the warm gray background to a soft sage green studio backdrop. Preserve the bottle shape, amber glass, ivory cap, label text, bottle position and pale stone surface. Keep the headline "DAILY CARE" exactly as it is, including its size and position. Preserve the light direction and shadow. Add no objects and make no other changes.

Compare the cap, label and bottle outline with the previous version. If any of them drift, return to the last accepted image and retry the background edit with a shorter instruction.

Step 3

Adapt it to a vertical post

Upload the accepted green-background version and switch the aspect ratio to 9:16 at 1K. This is a new composition task: ask the model to extend the scene and reframe the layout, so it has room to work in the taller canvas.

Adapt the uploaded skincare poster to a vertical 9:16 composition. Keep the same serum bottle, amber glass, ivory cap and "SERUM" label. Keep the sage green background, stone surface and light direction. Extend the background and surface to fill the taller frame. Place "DAILY CARE" near the top with a comfortable margin and position the whole bottle in the lower middle. Keep the bottle proportions natural. Add no new text or objects.

Check that the bottle has not stretched and that both the cap and base remain visible. Leave enough margin for the intended placement. For a bigger final version, choose a supported resolution and inspect the new output again.

Start with these GPT Image 2.5 settings

ControlStarting pointWhen to change it

Model mode

Flare

Start with Flare for fast iterations; try Sunburst when precise edits matter more than speed. Both support generation and editing, with quality set separately.

Model

GPT Image 2.5

Select image mode in the generator.

Quality

Medium

Choose Low, Medium, High, XHigh or Max. Start at Medium and increase quality when the image needs more detail.

Resolution

Use 2K or 4K when you need a larger output.

Aspect ratio

1:1

1K also offers 16:9, 2:3, 3:2 and 9:16. 2K offers 16:9, 1:1 and 9:16; 4K offers 16:9 and 9:16.

Reference

Optional

Upload an existing image when its subject or composition should guide the result.

How much does this three-step exercise cost?

Three 1K Medium generations cost 9 credits in either Flare or Sunburst. Retries and additional images cost extra.

Resolution / exampleCredits per image / totalQuality / calculation

GPT Image 2.5 · 1K

1.5 / 3 / 6 / 12 / 24

Low / Medium / High / XHigh / Max

GPT Image 2.5 · 2K / 4K

3 / 6 / 12 / 24 / 48

Low / Medium / High / XHigh / Max

Three 1K Medium generations

3 generations × 3 credits = 9 credits

If the result is close, fix the specific problem

What you seeWhat to changeInstruction to try

Misspelled or extra text

Reduce the copy to one short headline and keep its exact spelling in quotes.

Replace only the headline with "DAILY CARE". Remove all other headline text. Keep the product and layout unchanged.

The product changes during an edit

Use the last accepted image again and request only one change.

Change only the background color to sage green. Keep the bottle, cap, label, position and lighting unchanged.

The composition feels crowded

Decide which element is most important and give it more space.

Keep the headline at the top. Make the bottle slightly smaller and leave a clear gap between the two. Add nothing else.

The vertical version crops the product

Reframe the scene within the selected ratio and name the edges that must remain visible.

Show the complete bottle, including cap and base, with clear margins. Extend the background instead of stretching the product.

Frequently asked questions

What changed in GPT Image 2.5?

The update focuses on reference fidelity, more targeted editing and consistency through successive edits. Use those strengths for keeping a recognizable subject while changing its setting or presentation. Review the output rather than assuming every detail is preserved.

Can I edit my own photos?

Yes. Upload a reference image in LumiYing and describe the change. State what must stay the same, such as a face, product label or composition. For another edit, upload the chosen result as the next reference.

Should I start with High or 4K?

Start with the layout and prompt. Low or Medium can help you assess a draft; use a higher setting when the output needs it. A larger image does not fix an unclear instruction or guarantee correct text.

Does GPT Image 2.5 cost more than GPT Image 2?

GPT Image 2.5 and GPT Image 2 share the Low, Medium and High prices at 1K and 4K. At 2K, GPT Image 2.5 costs 3 / 6 / 12 credits, while GPT Image 2 costs 1.5 / 3 / 6. XHigh and Max are available only in GPT Image 2.5.

Put your next image brief to work

Choose GPT Image 2.5, start with one image, and refine the result with a clear edit instruction.