| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
|
|
|
| |
Simple algo. Use quad intrinsics to get neighboring pixels' (x & y)
positions in trochoid space. Compute tangent and bitangent from that.
Then normal as cross product. There's some artifacting on diagonal
boundaries.
|
| |
|
|
|
| |
Seems like the determinant of the jacobian is ~0, meaning it's not
invertible. That means the normals look fucked up.
|
| |
|
|
|
|
|
| |
attempt to fix hypotrochoid normals
key insight is that multiplying normal by transpose(invert(jacobian)) of
transform should work, but it fucking doesn't
|
| | |
|
| | |
|
| |
|
|
| |
Also overhaul how emitter planes are specified. Use normal and tangent.
|
| | |
|
| |
|
|
|
|
|
| |
* Use float param to select which gimmick to show
* Add colors & domain warping to gimmick 2
* Also optimize it. Colors let us avoid 4-fold neighbor check fanout
* Add rough/unoptimized version of gimmick 3
|
| |
|
|
|
|
|
|
|
|
|
| |
Included changes:
* Update LTCGI includes
* Add toggle to force world lighting
* Unity fucking sucks dick and balls and doesn't set LIGHTMAP_ON
consistently
* Add some more sdfs for raymarching
* Add separate LTCGI lighting mode for avatar vs. world
|
| |
|
|
|
|
| |
Debugging why shadowmap LTCGI emissions don't show up. This patch makes
the shader match the appearance of the LTCGI surface shaders much more
closely.
|
| | |
|
| |
|
|
|
|
|
| |
This is unconditional (for now). In VRChat, anything other than fully
wrapped vertex lighting usually looks bad.
Also add direct lighting to clearcoat.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
The old smooth_min(x, k) had the properties
smooth_min(1, k) = k
smooth_min(inf, k) = 1
Now it's smooth_min(x, k, j) with the properties
smooth_min(1, k, j) = k
smooth_min(inf, k, j) = j
|
| |
|
|
|
|
|
|
|
|
|
| |
Also:
* add gen_sdf script, which converts solid color (ideally b&w) decals
into sdf textures
* add 6 more decal slots
* add color, sdf options, alpha multiplier, tiling mode, and mask to
decals
* add token pasting macros (MERGE) to minimize code size of new decal
slots
|
| | |
|
| |
|
|
|
| |
Also rescale atlas so we can use texture sizes that are not a perfect
multiple of the grid size.
|
| |
|
|
|
|
|
|
| |
Use Third_Party/gen_atlas to create a 512x512 texture and put it in the
texture slot (the resolution is hardcoded to 512 in shader). Use sliders to
control the grid size. If desired, use global offset to make it possible
to linearly animate between e.g. 0-9 and have it display as ASCII
'0'-'9'.
|
| | |
|
| |
|
|
|
|
| |
Now it generates things in a fixed-sized grid which is derived from
charset's max codepoint. It does not interleave gaps, instead putting a
big gap at the end/bottom of the atlas.
|
| |
|
|
| |
Trying to beat claude 3.5 into submission. Little issues with placement.
|
| |
|
|
|
|
|
|
| |
I'm using an AAP (vrc.school/docs/Other/AAPs/) to create a frame
counter. That frame counter then drives this parameter. Thus the sparkly
pattern won't turn into gross flashing when the framerate is low.
Also add a speed parameter to control IGN.
|
| | |
|
| | |
|
| |
|
|
| |
Also add iq's "almost identity" curve.
|
| |
|
|
| |
Aces filmic is more accurate than just "aces".
|
| |
|
|
|
| |
Use a nice "almost identity" function to create a smooth transition to
the clamped value.
|
| |
|
|
| |
Looks better lol
|
| |
|
|
| |
Small perf win
|
| | |
|
| | |
|
| |
|
|
| |
Average LODs 9 and 11.
|
| |
|
|
|
| |
Drop fbm, it doesn't contribute as much visual interest per FLOP as
domain warping.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
| |
Uses iquilez's directional derivative model and a simplified lighting
model. Lighting is evaluated at each sample.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
Based on screen space uvs.
|
| |
|
|
|
|
| |
* Add third HSV slot
* Add ZTest enum
* Fix bug where SSR mask keyword can be set when SSR is disabled
|
| | |
|
| |
|
|
| |
Lower step sizes would gain energy.
|
