The combinatorial engine for AI-generated art

Stop regenerating.
Start combining.

Every AI tool answers "I need more variations" with "generate more." More credits, more GPU time, more randomness, more waiting. The combinator answers differently: the variations already exist inside your batch. You just need to extract and reassemble them.

Upload your AI-generated character images. Draw freeform shapes on the swappable regions — head, outfit, accessories. The combinatorial engine computes every possible assignment: face from image 1 + outfit from image 4 + accessories from image 9. Then face from image 2 + outfit from image 4 + accessories from image 7. Every permutation. 9 images × 3 shapes = 620+ unique character variations.

Pure combinatorial math Zero AI, zero GPU, zero credits Every pixel is yours

A combinator is not a blender.

When you search "combine AI images," every result shows you AI blending tools — neural networks that fuse two images into a new hybrid. That's a fundamentally different operation. Here's why it matters:

AI Blender (Midjourney /blend, Fotor, etc.)

Feeds your images into a neural network. The AI interprets both and generates a NEW hybrid. Output pixels are hallucinated — they don't exist in either source. Result is unpredictable. Costs GPU time and credits. Good for creative exploration. Bad for precise character work where you need exact control over what goes where.

Combinator (Compix)

Extracts exact pixel regions from your existing images using freeform shapes. Reassembles them in every mathematical combination. No neural network. The face from image 3 is the ACTUAL face from image 3 — not an AI approximation. Deterministic: run it twice, get the same result. No GPU, no credits, no hallucination.

The distinction matters because AI blending introduces drift — the very problem you're trying to solve. If your character's face drifted across 10 generations, blending two of them produces a face that looks like neither. Combinatorial reassembly gives you the EXACT face from the image where it looked best.

Every shape you draw multiplies your variations exponentially.

72
9 images × 1 shape
Just one region
620+
9 images × 3 shapes
Head + outfit + accessories
10,000
10 images × 4 shapes
Full character breakdown
100k+
10 images × 5 shapes
Every micro-region
The formula
sources × (2shapes - 1) × variantsactive_shapes
Each shape is an independent axis of variation. Each image is a source for every shape. The combinations are exhaustive — every possible assignment is computed.

The combinator breaks the generation loop.

Every AI image platform monetizes the generation loop. You need more variations → you generate more → you spend more credits. A combinator that produces 620+ variations from 9 existing images without generating is revenue they lose. They have no incentive to build it. Compix has no generation revenue to protect. That's why this exists here and nowhere else.

About the combinator

Character images generated from similar prompts — same character in different poses, outfits, or expressions. The images should have roughly the same framing and resolution. AI-generated character variants from Midjourney, Stable Diffusion, Flux, Leonardo, or ComfyUI are ideal.
Yes. Any set of aligned images with swappable regions works. Product photography (swap backgrounds, swap product colors), architectural renders (swap lighting, swap materials), game assets (swap textures, swap equipment). The combinator is region-agnostic — it works with any freeform shape you draw.
One anchor plus up to 50 comparison sources. With 50 images and 3 shapes, the number of possible combinations is astronomical. In practice, 5-15 images with 2-4 shapes produces a manageable and useful variation set.
No hard limit on the number of shapes. Each additional shape multiplies your combinations exponentially. In practice, 2-5 shapes covers most character variation needs (head, outfit, accessories, weapon, background).

The combinator is one piece of the workflow.

The Full Variation Guide

Step-by-step walkthrough of freeform region extraction, the compound math, and the complete workflow from source images to finished variations. See the guide →

Combine Parts From Different Images

The practical guide: take the face from image 1, the outfit from image 3, the boots from image 7. Every combination. How to combine →

Detect Drift First

Before combining, blink your batch to spot which regions drifted. The diff heatmap shows what's worth extracting. Detect drift →

The variations are already in your batch.

9 images. 3 shapes. 620+ combinations. Zero generation cost.

Open the Combinator — Free →