Gravitational-wave detectors such as LIGO, Virgo, and KAGRA are highly sensitive instruments susceptible to numerous noise sources. Short-duration transient noise events, known as glitches, pose a particular challenge for data analysis pipelines as they can mimic or obscure astrophysical signals. We present GlitchGAN, a class-conditional generative model, built upon the Conditional Derivative GAN (cDVGAN) architecture, that is capable of synthesizing glitches directly in the time domain. The model is trained on high-quality reconstructions of seven common glitch types observed during LIGO’s third observing run (O3): Blip, Fast Scattering, Koi Fish, Low-Frequency Burst, Scattered Light, Tomte, and Whistle. We show that GlitchGAN generalizes effectively, learning to reproduce a diverse glitch space that is consistent with these reconstructions. Moreover, because the model is conditioned on glitch class, it can generate or transitional glitch morphologies by interpolating across the class-conditioning vector after training. GlitchGAN generates 1000 glitches in under 22 seconds on a CPU, making it suitable for large-scale glitch synthesis for detector simulations, mock data challenges, and pipeline validation. Synthetic glitches are validated against real glitch reconstructions using the Gravity Spy classifier, widely used in the GW community for glitch classification, and an unsupervised analysis using UMAP embeddings. Gravity Spy classifies the majority of GlitchGAN’s synthetic glitches as the correct class while the UMAP analysis shows substantial overlap between real and synthetic samples in the reduced latent space. To investigate residual distributional differences between real glitches unseen during training and generated synthetic glitches, we train a separate holdout GlitchGAN model on nearly the same dataset but with a small held-out split, and use a downstream convolutional neural network to distinguish these held-out real glitches from synthetic glitches produced by the holdout model. While this classifier reliably detects synthetic samples in a clean, noise-free representation, this detectability drops substantially once real and synthetic glitches are injected into realistic detector noise, the condition under which synthetic glitches would typically be deployed for downstream applications. Despite remaining statistically distinguishable, we show that GlitchGAN-generated glitches are practically useful: augmenting real training datasets with synthetic samples matches simple duplication of the same real data when data is abundant, and increasingly outperforms it as real data becomes scarcer. Finally, we highlight a critical limitation of magnitude-only spectrograms: classifiers operating on magnitude $Q$-transforms can confidently misclassify physically unrealistic glitches from less robust models, underscoring the need for complementary validation methods that preserve phase information.