AI Art World
Dreaming in latent space
Diffusion Process
Watch an image emerge from pure Gaussian noise as the denoising timestep progresses from t=1000 to t=0.
q(xt | x0) = √ᾱt · x0 + √(1−ᾱt) · ε   where ε ~ N(0,I)
Timestep t
t = 800
CLIP & Semantic Space
CLIP (Contrastive Language–Image Pretraining) encodes text and images into a shared embedding space. Click nodes to explore semantic relationships.
Similarity: —
ControlNet Systems
ControlNet adds spatial conditioning to diffusion models — edge maps, depth, pose, segmentation, and normals guide composition without losing artistic freedom.
Conditioning Strength
0.75
LoRA & Style Blending
Low-Rank Adaptation fine-tunes a small set of weight matrices, injecting style with surgical precision. Blend multiple LoRAs — but watch the conflict meter.
Style Conflict Level
Low conflict
Sampler & CFG Lab
Different samplers trace distinct paths through latent space. CFG scale balances creativity vs. prompt adherence.
CFG Scale
7.0
Steps
20
AI Aesthetics & Ethics
Three dimensions of the AI art conversation: authorship, data provenance, and the spectrum from tribute to appropriation.

Authorship Spectrum

Pure HumanCollaborativePure Machine
Collaborative

Training Data

Scraped Web 65% Licensed 20% Synthetic 15%

Style Mimicry

Where does AI style use fall on the ethical spectrum?
TributeTransformationAppropriation
Tribute / Homage
AI Art Pioneers
Artists who have shaped the conversation between human creativity and machine intelligence — from 1970s algorithmic art to today's latent-space explorations.
Glossary
Essential vocabulary for navigating the world of AI-generated imagery.
Prompt Engineering Lab
Discover how token ordering, weights, and style keywords shape generative output. Click tokens to see their influence, adjust weights, and explore prompt anatomy.
Active Token Weight
1.00
Prompt Complexity
5 tokens
Embedding(prompt) = Σ w_i · token_i  |  CLIP encodes full sequence with cross-attention
VAE Latent Explorer
The Variational Autoencoder compresses images into a 4×64×64 latent space. Explore how interpolation through latent space creates smooth transitions between concepts.
Interpolation α
0.50
The VAE encodes images as distributions (μ, σ) in latent space. Sampling from these distributions introduces controlled variation — the source of generative diversity.
z = μ + σ · ε   where ε ~ N(0,I)