MaterialX Trainer

Fit a MaterialX material into a neural texture compression (.ntc) model, right in the browser.

Source

Built-in MaterialX

loading default — or drag & drop a .mtlx file here

Bake resolution

Network (grid + MLP)

Preset

Feature levels

Finest grid res

MLP hidden width

MLP hidden layers

MLP hidden activation

Positional encoding

Dual grid G0/G1

Training

Batch size

Iterations: 10000

Learning rate: 0.010

Quantization

Object

Shape

View

Sampling

One decoded texel per sample.

LOD bias (force finer): 1.00

Status

Ready.

Model Info

Name: Untitled
Encoded payload: 24.1 KiB (24,718 bytes)
Runtime estimate: 175.8 KiB (179,996 bytes)
Latent grids: 4 (G0: 128×128×4 @ 2-bit, G0: 32×32×4 @ 2-bit, G1: 64×64×4 @ 2-bit, G1: 16×16×4 @ 2-bit)
MLP layers: 3 (33→16→16→39)
Decoder: 33 inputs · 1,479 parameters
MLP work per sample: 2.9K FLOP · 1 evaluation

Channels (0)

Channel	Source

Training loss

No training data yet.