Edit your image with Instruct: make smart changes using text

Instruct lets you edit an existing image using natural language. You can add one or more reference images, depending on the model, and describe what you want to change. The system will generate a new version based on your instruction.

You can use it to change the color of an outfit, remove an object, adjust lighting, switch the season, or make many other types of edits. To change only specific areas of an image with masks, use the Editing tab instead: see Edit images with exactly 1.0 edit.

How to use it

Option 1 – Upload an image from your device

1. Choose a model that supports Instruct (see the list below)

2. Add one or more images with the + tile at the start of the image row, or drag them onto the page or paste them. On a model that supports Instruct, each new image arrives as an Instruct reference

3. Click an image's thumbnail to check its settings: Role should be Reference and Method should be Instruct. If the image arrived with another role, set Role to Reference and pick Instruct in the Method menu

4. Write your instruction in the prompt field

5. Click the round arrow button to generate

Option 2 – Use an image from your project

1. Hover over an image in your project

2. Click the Remix button that appears

3. The image is added to the prompt bar

4. Make sure the model in the prompt bar supports Instruct (see the list below), change it if needed, and check the image's settings as in Option 1

5. Write your instruction in the prompt

6. Click the round arrow button to generate

Referencing images with @-mentions

When you have one or more Instruct images attached, you can refer to them directly inside your prompt using @-mentions. Type @ in the prompt field and pick an image from the list.

Example:

Move the model from @Image-1 to the office in @Image-3. Put glasses on the model from @Image-2

• Instruct images are named automatically (Image-1, Image-2, and so on), and the name shows on the thumbnail. To rename one, click its thumbnail and change Name: letters, digits, - and _, up to 25 characters.

• Only Instruct images can be mentioned and renamed. Images used for Image to Prompt, Sketch, Scene or a pin have no name; their thumbnail shows how they are used instead.

• You can swap the image inside a reference without losing the mention: click the thumbnail and use Replace image…. Useful when iterating on the same slot with different visuals.

How it works

The uploaded or selected images become the reference for generation. Your text instruction is used to make targeted changes to that image.

The Method menu in an image's settings offers only what the selected model supports. Methods the model doesn't support are greyed out, and hovering one tells you so.

As of October 2026, these Base Models support Instruct: FLUX 3 Image, Flux.1 Kontext, GPT Image 2, GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, Ideogram 4.5, Nano Banana 2, Nano Banana 2.1, Nano Banana Pro, Qwen Image 2.1 Pro, Recraft V4 Styles, Riverflow 2.5 Pro, Seedream 5.0 Flash and Seedream 5.0 Pro. Artist Styles, Private Styles, Exactly 1.0 Agentic and video models don't. New models are added often, so check the Method menu: Instruct is greyed out where it isn't supported.

The prompt bar holds up to 10 images, and some models accept fewer Instruct images: Flux.1 Kontext takes two and Ideogram 4.5 takes five. Images over the model's limit are dimmed and left out of the request.

Instruct images can be combined with each other and with Image to Prompt images. Sketch, Scene, Character pin and Product pin can't be used in the same request: whichever you add later is dimmed and left out.

In the Editing tab, every image you add after the Editing image works as an Instruct reference, so there is no method to choose.

To keep the shape of your reference image, choose Auto in the aspect ratio menu. Models that support it offer Auto while an Instruct image is attached. With a fixed ratio, the result is generated in that ratio, so the composition can change.