gene46 gives everyone the same starting conditions: a random initial population and a blank canvas of evolutionary possibility. Yet after 50 generations, two players' results can look completely different. The difference lies entirely in their selection strategy β how they swipe, what they value, and when they fork.
This guide shares proven strategies for efficiently cultivating the most beautiful, complex generative art your evolution can produce.
1. Define Your Aesthetic Before You Start
The single most important factor in evolution quality is consistent selection pressure. Before you begin swiping, decide on a clear aesthetic goal:
- "I want cool blue and purple tones"
- "I prefer bold, thick geometric lines"
- "I want intricate, dense mosaic patterns"
- "I'm aiming for minimalist symmetry"
Write it down if you need to. Then apply that criterion consistently across every swipe. Inconsistency β liking warm colors on Monday and cool colors on Tuesday β creates mixed selection pressure that slows or reverses progress.
2. Early Generations: Be Ruthless but Fast
In generations 1β5, patterns are essentially random. Don't overthink individual cards β they're all noise. Instead:
- Swipe quickly based on gut reaction.
- Eliminate anything obviously "wrong" for your theme (wrong color family, wrong structural type).
- Right-swipe anything that has even a faint hint of your desired direction.
Aim to right-swipe 5β10 individuals per generation in these early rounds. Too many kept individuals (20+) dilutes selection pressure; too few (1β2) reduces genetic diversity and risks a genetic bottleneck.
3. Mid Evolution (Gen 5β20): Slow Down and Select
Around Generation 5β10, genuine patterns begin emerging. This is where your selection strategy really matters. Slow down and evaluate each card:
- Look for the specific visual properties you defined (color, shape, density).
- Eliminate anything that violates your aesthetic, even if it's "interesting."
- Right-swipe individuals that show clear progress toward your vision.
Optimal Selection Count per Generation
Research on Interactive GAs suggests keeping 5β12 individuals per generation balances selection pressure (driving evolution) and genetic diversity (preventing stagnation):
- 1β3 kept: Fast convergence, but high risk of genetic bottleneck. Works well if you want extremely focused evolution.
- 5β12 kept: Balanced. Recommended for most goals.
- 15+ kept: Slow evolution, but broad diversity. Good for exploring unusual territory.
4. Use Fork Strategically
The Fork feature is gene46's most powerful tool. Fork creates a copy of a thread at its current state, allowing you to evolve two independent lineages from the same ancestor.
Best times to fork:
- When you've reached a beautiful generation and want to explore two different directions from it.
- Before trying an experimental selection strategy that might damage a promising lineage.
- When evolution is stuck β fork and apply radically different selection in one branch.
5. Mosaic vs. Line Art: Different Breeding Strategies
Line Art Mode
DNA encodes BΓ©zier curve parameters. Evolution trends toward elegant, flowing geometric curves. Selection strategy: prioritize individuals with pleasing curve shapes and color gradients. Stagnation usually resolves within 3β5 generations due to the continuous nature of curve parameters.
Mosaic Mode
DNA encodes a CPPN (Compositional Pattern Producing Network) β a mathematical function that maps pixel coordinates to colors. This creates fractal-like, symmetric, complex patterns. Selection strategy: look for structural coherence and interesting repetition patterns. Mosaic evolution can stagnate more easily β fork frequently and maintain high diversity.
6. Recognizing & Escaping Stagnation
Signs your evolution is stagnating:
- Every card in a generation looks nearly identical.
- You struggle to find even 3β4 cards worth keeping.
- Visual quality is decreasing generation-over-generation.
Recovery strategies:
- Fork and apply maximum diversity selection (keep unusual outliers).
- Temporarily change your selection criterion to force a different direction.
- Keep a lower count (2β3) to create stronger selection pressure.
About the Author
gene46 Development Team
Building interactive evolutionary generative art at the intersection of game design and Genetic Algorithm research. These tips come from thousands of hours of internal playtesting.