2-D diffusion
Training
Sampling
Traces
Inpainting
Dataset
Dataset
Dataset
Moons
Spiral
Circles
Blobs
Sine wave
Checkerboard
Hello world
Load CSV
Dataset size
Seed
Add points
Diffusion
Diffusion
Schedule
VP
VE
SubVP
Cosine
Linear
Prediction
Noise ε
Clean data x₀
Score
Flow matching (velocity)
Objective
Continuous
Discrete
Weighting
Simple
VLB
Flow matching
Sampler
Sampler
Sampler
Ancestral
Predictor–corrector
Euler–Maruyama (SDE)
DDIM
Forward diffusion
Euler (ODE)
Heun (ODE)
Forward ODE
Blended inpainting
Steps
Samples
Grid resolution
Training
Training
Optimizer
Optimizer
Adam
SGD
Momentum
Learning rate
Batch size
Momentum
Scaling
Weight decay
Schedule
Warmup steps
LR half-life
Minimum LR ratio
Next-step LR
0.00100
Display
Display
Layers
Density
Contours
Data
Traces
True score
Estimated score
True score grid
Estimated score grid
Likelihood
Show model likelihood
Compute likelihood
Method
Cached particle KDE
Probability-flow ODE
Solver
Euler
Heun
RK4
Steps
Particles
Bandwidth ×
Resolution
Plot
Plot settings…
PNG scale
Export PNG
Refresh
Network refresh
Every frame
30 fps
15 fps
5 fps
Data density
True score
Estimated score
Fit
Loading...
Schedule plot
Signal-to-noise (log scale)
Log signal-to-noise
σ · standard deviation
σ² · variance
α · scale
f · drift coefficient
g · diffusion coefficient
Viewed t = 0.250
-10
-5
0
5
10
0
Time
1
Train
Step
0
Loss
—