How to retain full quality on a rotated textured quad?

[x4000]

So, I'm rendering a textured quad via Graphics.DrawMesh, on an orthographic camera, for windows/mac standalone, in the latest unity 3 beta. All is well when there are no rotations, I have absolutely pixel-perfect crisp graphics.

However, when I rotate the textured quads, it quickly starts to look rather horrible in terms of quality. In SlimDX, running over top of the Direct3DX sprite extensions, it pretty much handled this for me -- there were some artifacts, but they were minor to the point of only being noticeable as the rotation angle changed.

On the Unity side, I've tried:

1. Using Bilinear filtering for the Texture2Ds that are rotated (normally it is Point for maximum correctness/quality when not rotated), but that makes the entire texture look fuzzy, despite the fact that the edges look better.
2. Trilinear, which does the same as Bilinear.
3. Adjusting overall quality settings, including forcing on aniso, 8x multisampling (from 16x, that's the highest my card will go) -- and none of that has any effect whatsoever.

I've been all through the Unity documentation a gazillion times for this -- first for my game Tidalis back four or five months ago, and now for my other game AI War, which I'm now porting to Unity.

Surely there's something I'm just missing...?

First Edit:

Here's the direct image I'm using as the texture on the textured quad, if anyone else wants to check out what is happening:

Second Edit:

Here's an image that it might be easier to see the general bilinear distortion on, because of better contrast:

This is the sort of quality drop I'm seeing, and I'm seeing it in your image, too:

See how that one starship, which has been rotated with bilnear filtering applied, looks all blurry while the non-rotated point-filtered sprites all look perfectly clear? That's the crux of the issue I'm running into here. Aside from the rotated starship looking pretty blurry in general, worst of all that pink border is incredibly blurry, which makes it stand out like a sore thumb near its un-rotated-point-based peers.

Certainly that can be bypassed by making everything bilinear filtered, but that looks pretty wretched, too. Here's an example of the quality differences in general between bilinear and non-bilinear for non-rotated sprites:

Nice and crisp, no bilinear filtering except on the rotated missiles:

Unfortunately blurry, with bilinear filtering on everything:

This is really quite a challenge!

[Eric5h5]

Here's my test in Unity. When not rotated, I can't see any difference between point and bilinear filtering, so there doesn't seem to be any reason to use point filtering here, which can never work well for rotated sprites. I'm not sure what rotations you used, so I went with 22. It's a bit hard to tell with such tiny images, but maybe it looks better than what you're getting?

As for what doesn't work:

Trilinear filtering: this blends between mip levels, which you don't need for 2D, so it will have no effect.

Anisotropic filtering: this reduces blurriness of far textures when viewed at oblique angles, which again will have zero effect in a 2D game.

FSAA: this antialiases edges of polygons, so it can never have any effect on texture quality, no matter how high you set it.

If my image is better than what you're getting, make sure:

The orthographic camera and quad positioning is set up properly, so you don't have texture pixels disappearing "in between" the pixel grid.

You're using RGBA uncompressed textures, not DXT5.

Edit: There really isn't any difference between point and bilinear when the sprite is positioned correctly. I took screenshots of the bigger sprite and blew it up 300%; they are identical. Then I took screenshots of the quad when positioned at (.5, .5) so that the texture pixels are halfway between the pixel grid. Clearly, the only way to get a true pixel-perfect display is when texture pixels are exactly aligned to the pixel grid. Bilinear is blurry; point is clearer, but still not perfect (some of the pixels are doubled). You can get nice sub-pixel movement with bilinear though, at the cost of blurriness when between the pixel grid.

The artifacts when rotated are simply the result of working with pixels instead of vector images. There's no way to do it perfectly...the bottom row is rotated at 45 when using Photoshop, then bilinear filtering, then point. I'd say bilinear actually gets the best results here.