How to Use GPT Image 2.5 on Movmix

GPT Image 2.5 — the same family ChatGPT ships as Images 2.5 — landed on 8 September 2026. Officially it is faster than 2.0, better at light and surface, and less likely to replace the subject when you edit from a reference.

On Movmix it lives under Image. This walkthrough covers text to image and image to image only. Do not look for it in Video. It is not there.

Open Image and lock the model

Go to movmix.ai and click Image in the left rail. Skip the Dance and Glamour cards on the home page. Those run effect templates, not this generator.

In the model list choose GPT Image 2.5. If the list is split:

Start on Flare. Switch to Sunburst when the composition is already right.

Use 4:5 or 3:4. A tall kitchen still reads as a cover. It also stays clear of the wide street portraits in older posts.

Text to image: name the object, the surface, the light

GPT Image 2.5 follows materials if you keep the prompt short.

Leave out cinematic, 8k, masterpiece, ultra detailed. Those words push a catalog kitchen toward a poster.

Image to image: upload that file, change one fact

Do not roll a new text-to-image and hope the kettle matches. Upload the still you already accepted.

One variable per pass. Move the cup and recolor the steel in the same line, and you will not know which instruction slipped.

The same rule holds for a face. Upload the portrait. Ask for a new room. Tell it not to touch the face. That is the point of 2.5: hold the subject, change the set.

If the still needs type, write one short line

GPT Image 2.5 is steadier with letters on a surface than 2.0 was. It is still a label tool, not a layout tool.

One line, one place. A paragraph on the card will collapse. Short pack copy is fine; body text belongs in a design app.

What usually breaks the first session

• Hunting for 2.5 inside an Inspiration template. It sits in the Image model list.  
• Opening with “a beautiful kitchen, lots of atmosphere.” The kettle becomes a showroom prop.  
• Generating three near-copies from text and picking the least wrong one. Run image to image from the keeper.  
• Pasting a ChatGPT `@Sketch` workflow into Movmix. There is no sketch canvas here. Composition comes from the prompt and the reference file.

Next step

Open Image → select GPT Image 2.5 → text to image once → image to image on that same file.

That is the whole use of this model on Movmix. Faces and longer series can wait for a later post.