Taste in generation

Once your Aesthetic Profile is unlocked, you can activate it during image creation through the Memory dropdown in the canvas chat toolbar. Memory lets Virse apply your learned preferences to every generation — so outputs feel like yours without needing to describe your style in every prompt.

Memory modes

The Memory dropdown offers two modes:

Mode	Source	What it does
Aesthetics	Your Aesthetic Profile (radar chart)	Applies your six-dimensional preference scores to guide generation
Style	A trained style model	Applies a specific visual style learned from a set of reference images

You can enable Aesthetics or Style, but not both at the same time. Enabling one automatically disables the other.

Using Aesthetics

Toggle Aesthetics on in the Memory dropdown. Your radar chart is displayed alongside the toggle so you can see the profile being applied.

With Aesthetics active, your six dimension scores (Vivid, Complexity, Order, Realism, Abstract, Temporal) are passed to the generation pipeline alongside your prompt. The result is output that reflects your personal taste — more saturated if you lean vivid, more minimal if you lean toward simplicity, and so on.

Training a Style model

Style models let you go beyond your general aesthetic profile and teach Virse a specific visual language from reference images.

Creating a new Style

1. Open the Train panel from the workspace.

2. Select images — Upload new images (drag-and-drop or file picker) or choose from your existing asset folders. You need between 1 and 20 images.

3. Configure — Give your model a name and select what to optimize for (currently Style).

4. Submit — Virse trains the model in the background. You'll see a progress indicator on the model card.

Training is asynchronous — you can continue working on the canvas while it completes.

Managing your models

Your trained models appear in the Train panel, organized by category:

• Personal — Models you've created

• Team — Models shared with your organization

• Shared With Me — Models others have shared with you

Each model card shows thumbnail previews from its training images and a status indicator (in progress, completed, or failed).

Re-training a model

Click any completed model to view its details. From there you can:

• Add new reference images or remove existing ones

• Submit the updated selection to re-train the model

• The model re-trains with the new image set, refining or redirecting its visual style

Using a Style in generation

1. Open the Memory dropdown in the canvas chat.

2. Toggle Style on and select your trained model.

3. Generate as usual — your prompt is now influenced by the visual language of your style model.

How it all fits together

Aesthetics and Style serve complementary roles:

• Aesthetics is your broad creative fingerprint — it nudges every generation toward your general taste.

• Style is a focused lens — it applies a specific trained visual language to a generation.

Both are optional. You can generate without Memory entirely, use one mode, or switch between them as your creative needs change.