First, some background on why this occurs.

Without mipmaps, as you shrink your texture down, the tiny fringe of half-coloured texels providing you with a pleasant anti-aliased border between two colored areas shrinks to nearly nothing. The probability that any specific pixel’s texture pattern lands proper on that razor edge goes towards zero, and so most pixels find yourself trying up their texture from one aspect or the opposite, 100% the color on the left or 100% the color on the suitable, leading to that harsh stairstep transition with no mixing.

Auto-generated mipmaps assist with this by pre-generating smaller variations of the asset which might be pre-blended for quick lookups (and higher cache coherency which is nice for efficiency). But once they solely have a raster picture to go off of, they need to make guesses concerning the true steady color discipline they’re alleged to be approximating. Usually they will use a easy field filter or some barely extra refined methods of averaging the colors already baked into your raster. That averaging tends to smear out your borders greater than a real supersample of the supply vector would, contributing to the slight fuzziness you see.

You can enhance this considerably by making your personal mipmaps immediately out of your vector supply. Your vector enhancing software program is ready to apply supersampling to the bottom reality steady picture to get a really high-quality downsample, which something working from simply the raster can solely approximate.

So as a substitute of exporting only a single high-res sprite, export an entire sequence, every half the dimensions of the earlier. You can mix these photos right into a mipmap chain in DDS texture format — there are plugins for Photoshop and different picture instruments to assist with this, or you need to use DirectX utilities to do it manually.

Importing this into Unity, it would use your mipmaps as a substitute of guessing at its personal with recursive filters. Then anytime your sprite is displayed at precisely one of many resolutions you exported, and aligned with the pixel grid of the display screen, you will get crisp edges identical to your direct vector output. So for very best quality, attempt to export your sprite at a decision such that the standard sizes it seems in your sport every carefully match the decision of one of many supply photos within the mip chain. You may even swap between totally different variations of the sprites for closeup and much photographs, if the sizes you want do not align nicely with an influence of two relationship.

But in-between these sizes you may nonetheless run into bother. By default, utilizing trilinear filtering, the GPU will common two close by mip ranges to estimate the suitable color for every pixel at an in-between measurement. This can once more introduce just a little fuzziness, each from the averaging, and since a kind of ranges is just too low-res/blurred for the dimensions you are drawing at. You can power the filtering to bilinear / select simply the closest mipmap to pattern from, which can introduce just a little upscaling fuzz when scaled up and just a little aliasing when scaled down from the closest native measurement you saved — and a visual pop throughout scaling because it switches sources discretely. Or you may apply a mipmapping bias to shift the lookup just a little extra strongly towards the extra detailed mip, buying and selling just a little further aliasing for crispness, whereas preserving the transitions clean and pop-free.