How to change between image models in Assist (and what each one does)

Assist supports a few different image-generation models, and each one is better at different things. The cleanest way to work with them is to build a small set of image-gen templates — one per model — so you can switch by simply picking a different template. This article shows you how to do exactly that, and what each model is good for.

What you'll learn

• How image generation works inside Assist
• What each of the four image model options does
• A quick rule of thumb for picking the right model
• How to change the image model on any image-gen template
• The recommended setup: one template per model

How image generation works in Assist

Image generation in Assist runs through image-gen templates. A template is just a reusable setup — name, description, settings — that you launch from chat whenever you want to generate images that way.

Every image-gen template has an Image Generation Model setting. That setting decides which AI model is used to create the image. Change the setting, and the same template can produce very different results.

The four image model options

When you open the Image Generation Model dropdown inside a template, you'll see four choices:

Inherit from organisation (default)

Not a model itself — it just uses whatever model your organisation has set as the default. Good if you don't want to think about it. If your admin changes the org default later, the template automatically follows along.

Gemini 3 Pro Image

Google's flagship image model. Best for imagery — photographic realism, lifestyle shots, scenery, product photography, illustrative scenes. Higher quality but slower and more expensive per generation.

Gemini 3.1 Flash Image

The faster, cheaper sibling of Gemini 3 Pro. Same family, lighter weight. Great when you need to iterate quickly or generate in volume and don't need final-grade polish.

GPT Image 2

OpenAI's image model. Best for text and graphical elements — posters, ads, UI mocks, logos with type, signage, infographics. It's the one to use whenever text inside the image needs to come out legibly.

Quick rule of thumb

How to change the image model on a template

1. Open Library from the left-hand nav
2. Click + New Template (or open an existing image-gen template you want to edit)
3. Fill in the template details — name, description, and so on
4. Scroll down to Image Output Settings
5. Open the Image Generation Model dropdown
6. Pick the model you want — Inherit from organisation, Gemini 3 Pro Image, Gemini 3.1 Flash Image, or GPT Image 2
7. Click Create Template (or Save if you're editing)

That's it. The template is now locked to that model until you change it again.

The recommended workflow: one template per model

If you do any serious image work, the cleanest setup is to build three templates — one per model:

• One using GPT Image 2 for graphics and text-heavy work
• One using Gemini 3 Pro Image for high-quality imagery
• One using Gemini 3.1 Flash Image for quick iteration and volume

Once they exist, you switch models by picking a different template — no fiddling with settings mid-project, no second-guessing which model is on.

Worked example

You work in marketing. You need to make a lot of images, of different kinds. So you set up three templates in your Library:

• Branded Posters — image model set to GPT Image 2. You use this for anything with copy on it: social posts, poster mockups, in-product graphics.
• Hero Imagery — image model set to Gemini 3 Pro Image. You use this for campaign hero shots and lifestyle photography.
• Quick Mood Boards — image model set to Gemini 3.1 Flash Image. You use this when you're exploring ideas and need a lot of options fast.

Next time a brief lands, you don't have to think about which model to use. You just pick the right template for the job.

Using your templates from chat

Once your templates are saved, using them is easy:

1. Go to the main chat
2. Click Library (or search via the chat input)
3. Type in the template name
4. Select it — you're ready to generate

Tips & common pitfalls

My image has wonky or unreadable text.

You're probably using a Gemini model. Switch to GPT Image 2 for anything with text — it's much better at rendering letterforms cleanly.

My photos look too "graphic" or stylised.

Try Gemini 3 Pro Image. It's tuned for photographic realism, lighting, and texture, where GPT Image 2 leans more illustrative.

My generations are slow and I'm just exploring ideas.

Switch to Gemini 3.1 Flash Image. It's the same family as Gemini 3 Pro but much faster and cheaper — perfect for iteration rounds.

My org default changed and now my templates are producing different results.

Templates set to Inherit from organisation follow the org-wide default. If you want a template locked to a specific model, pick that model explicitly in the dropdown rather than leaving it on Inherit.

I can't see my template in chat.

If you (or a teammate) can't find it, it may still be in your Private space. Promote it to Workspace or Organisation level so it shows up for everyone who needs it.

Build them once, and you'll never have to think about which image model to use again.