# Matrix → Crop — variations from one generation

**Matrix turns one generation into several variations.** Instead of paying for N
separate generations, the model returns a single image tiled with N versions of
your subject — one generation's credits for the whole set. The crop tool then
splits that image along the same grid into separate full-resolution nodes, for free.

1. **Turn on Matrix.** In the chat bar on **Image** (text-to-image or
   image-to-image), toggle **Matrix** on. Pick a **Composition** — `2×2` (four),
   side-by-side or stacked (two), or triple side-by-side / stacked (three) — and a
   **Creativity** level (Subtle / Balanced / Wild) for how far the variations
   drift. The model decides *what* to vary from your prompt.
2. **Generate.** One image lands holding all the variations — a single
   generation's credits, not one per variation.
3. **Open Crop** on that node and pick the grid preset that matches your layout:
   `2×2` → **2×2**, side-by-side → **1×2**, stacked → **2×1**, triple side-by-side
   → **1×3**, triple stacked → **3×1**. The splits fall exactly where the matrix tiled.
4. **Keep the winners.** Click individual cells, or **Select all**, then **Crop** —
   each selected variation becomes its own full-resolution image node on the canvas.

Note: you can re-pick the Composition and Creativity from the matrix pill before
generating. Because Crop is a free local operation, sweeping wide with Matrix and
culling with Crop costs the same as a single generation. See
[Chat bar: generating images](/docs/product/chat-image-gen), [Crop](/docs/product/image-tools),
and the [prompt-polish batch playbook](/docs/playbooks/prompt-polish-batch).