This project is inspired by @nakakoutv and @yukku 's work on stamping pixel grids. The idea is to stamp an entire 4x4 pixel grid all at once by simply setting the costume and whirl as stored in a LUT for every value from 2^16. This improves upon @nakakoutvs by not having direction or fisheye (faster) whilst still retaining a small number of costumes. I've written some quite complex C++ code based on @yukku's idea of a simulate annealing program that can find a near optimal covering of a target number of costumes with the pixel values, required for this to work. The algorithm works like follows: 1. Init target costume number of costumes as black 2. Create a lut table across all whirl/fisheye/direction effects that store for each output pixel what the source pixels were. 3. Find the union across the source pixels, this is our modulation set. 4. Begin the simulated annealing algorithm where we will flip bits of the costume randomly, and each time we check if the number of unique 4x4 outputs produced by applying every effect combo in the LUT goes up. if the temperature is high we can accept even if this reduces the number of unique 4x4 outputs. The lower the temperature the less chance of accepting a worse output. always accept a better output. 5. When choosing bits to flip we always pick from the modulation set and we pick a random costume each time. 6. We stop once our set of costumes covers all 65536 possible 4x4 outputs. 7. If a valid covering isn't found reuse the partial costumes as the starting point for another round of SA Credit for concept: @yukku's: https://scratch.mit.edu/projects/1289776074/ @nakakoutv's: https://scratch.mit.edu/projects/1306212839/ Update 22/04: replaced costumes with new mobile friendly versions that won't glitch on android mobiles
