Project MonetRequest demo
Home/Blog/How to Use ChatGPT Sketch & Images 2.5 Editing Tools

Creator Tools · Project Monet Briefing

How to Use ChatGPT Sketch and Images 2.5 Editing Tools

ChatGPT Images 2.5 adds more than a model upgrade. OpenAI also introduced Sketch, templates, image comments and prompt sharing, while improving multi-turn editing so earlier changes are more likely to remain stable during later revisions.

Published 2026-09-14 · Updated 2026-09-14 · By Project Monet Editorial Team

How to Use ChatGPT Sketch and Images 2.5 Editing Tools — Project Monet editorial graphic

01

Overview

ChatGPT Images 2.5 adds more than a model upgrade. OpenAI also introduced Sketch, templates, image comments and prompt sharing, while improving multi-turn editing so earlier changes are more likely to remain stable during later revisions.

This guide focuses on the creator workflow rather than the API.

02

1. Start with Images in ChatGPT

OpenAI says ChatGPT Images is available on web, iOS and Android across all tiers. You can ask for an image in a normal conversation or open the Images experience. You can also upload an existing image and describe the change you want.

If a specific new interface element is missing, remember that rollout can be staged even when a feature is announced for all tiers.

03

2. Use Sketch when words are not enough

Sketch lets you draw a rough visual reference inside ChatGPT. OpenAI says you can invoke it by typing @Sketch on supported mobile experiences and then draw the idea you want the model to interpret.

The sketch does not need to be polished. Its job is to communicate structure: where an object should sit, the contour of a product, the rough layout of a poster, the placement of a person or the shape of an outfit.

After drawing, add a short text instruction describing the intended style, materials, lighting, colors and any details the rough sketch cannot communicate.

04

3. Separate structure from style

A reliable workflow is to use the sketch for geometry and placement, then use text for visual direction. For example: sketch the position of the product and headline, then describe 'warm editorial lighting, off-white background, minimal luxury product photography, leave negative space around the headline.'

This reduces the amount of spatial reasoning the text prompt needs to carry alone.

05

4. Use templates for common formats

OpenAI introduced templates for popular image types such as posters, merchandise and product photos. Templates are useful when the job starts from a known format rather than a blank canvas.

Treat the template as a starting structure, not a guarantee that every brand rule is automatically preserved. Add explicit constraints for typography, logo placement, product identity, background, aspect ratio and any element that cannot drift.

06

5. Edit one thing at a time

Images 2.5 is designed to make more focused edits while preserving the rest of the image. Take advantage of that by making narrow revision requests: 'change only the background to warm beige,' 'replace the bottle cap with matte black,' or 'move the headline higher without changing the product.'

OpenAI says the model is better at keeping previous changes stable across multiple turns. That does not eliminate drift completely, so verify the image after every important revision.

07

6. Use image comments for targeted revisions

On supported mobile interfaces, OpenAI now allows comments on generated images. Comments are useful when the problem is local and hard to describe globally. Point to the relevant area and explain exactly what should change.

For sensitive production work, compare the edited image to the previous approved version before accepting it. Check faces, product geometry, logos, text, shadows and small brand details.

08

7. Ask for transparent backgrounds when needed

OpenAI says Images 2.5 improves complex layouts including transparent backgrounds. For design workflows, request transparency explicitly and verify the exported asset rather than assuming the background is truly transparent from how it appears in the chat.

This is useful for product cutouts, stickers, presentation assets, thumbnails and graphics that will be composed in another design tool.

09

8. Use multi-turn editing instead of restarting too early

One of the main 2.5 improvements is better consistency across repeated edits. If the base image is close, continue refining it rather than regenerating from scratch after every small issue.

However, if the composition has fundamentally failed, a fresh generation can still be better than stacking many corrective prompts. The practical rule is simple: preserve a strong base, restart a weak one.

10

9. Share prompts when the workflow is reusable

OpenAI now lets users share a prompt with a generated image. For teams, this can turn a successful creative direction into a reusable starting point. A teammate can reuse the idea with different subjects or details.

Do not treat a shared prompt as a complete brand system. Store important rules—logo usage, forbidden changes, colors, safe zones and legal copy—separately and include them when needed.

11

10. Build a creator-quality review checklist

Before exporting a final asset, inspect:

  • subject likeness and identity
  • product shape and labels
  • text spelling and hierarchy
  • logo accuracy
  • background transparency
  • shadows and reflections
  • hands and small objects
  • aspect ratio and crop safety
  • consistency with earlier approved edits

For ads and commercial content, also check usage rights and required disclosures.

12

What Images 2.5 improves—and what it does not guarantee

OpenAI says Images 2.5 improves subject preservation, editing precision, visual instruction following and multi-turn consistency. Those are vendor claims and not guarantees that every edit will be perfect.

The safest workflow is still human review. AI image tools can alter small details, text, identities or products in ways that are easy to miss at first glance.

13

FAQ

Where is ChatGPT Sketch?

OpenAI introduced Sketch with Images 2.5. On supported mobile experiences, type @Sketch and select the Sketch option. Rollout can vary by account and platform.

Can I upload my own image and edit it?

Yes. OpenAI’s Help Center documents uploading an existing image and describing the change you want.

Can ChatGPT edit only one part of an image?

Images 2.5 is specifically designed for more precise focused edits. You can also use selection/comment tools where available to target an area.

Can I create transparent-background images?

OpenAI says Images 2.5 improves handling of transparent backgrounds. Verify the final exported file before using it in production.

Are templates available on every interface?

OpenAI announced templates as part of the 2.5 rollout, but interface availability can still differ during rollout. Check the current Images UI on your account.

14

Sources

Recheck OpenAI’s Images Help Center and release notes before publication because interface details can change after launch.

Sources

Primary and supporting sources

Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.

Project Monet

Useful signals. Clear decisions. Better digital work.

Project Monet turns relevant shifts in AI, creator tools and the web into practical context—and builds focused websites for businesses ready to grow.

Request a free homepage concept