Rodziny modeli generatywnych

Generative model families · id generative

Różne sposoby uczenia rozkładu i generowania próbek.

Different ways of learning a distribution and sampling from it.

In one sentence

To różne rodziny modeli uczących się na przykładach, żeby tworzyć nowe obrazy, dźwięki lub inne dane.

These model families learn from examples to generate new images, sounds or other data.

Problem it solves

Potrzebujesz nowych wariantów treści, a samo rozpoznawanie istniejących przykładów ich nie przygotuje.

You need new content variations, and recognising existing examples alone cannot produce them.

What it does

Modele poznają wzory danych, a potem tworzą próbki przez odszumianie, przekształcanie losowych liczb, dekodowanie lub współpracę rywalizujących sieci.

Models learn data patterns, then generate samples through denoising, transforming random numbers, decoding or interacting networks with competing objectives.

Everyday example

Tworzysz kilka wzorów tapety, a różne generatory proponują nowe układy na podstawie poznanych obrazów.

You create wallpaper designs, and different generators propose new layouts based on images they have learned from.

This concerns you if you…

• chcesz zrozumieć różnice między generatorami
• potrzebujesz tworzyć nowe warianty danych
• want to understand differences between generators
• need to create new data variations

Examples

• Modele autoregresyjne · diffusion · VAE · GAN
• Autoregressive · diffusion · VAE · GAN

Subtopics

• Autoenkodery wariacyjne (VAE) (Variational autoencoders (VAE)): A model with an encoder that compresses data into a latent distribution and a decoder that reconstructs it from samples; trained by maximising the evidence lower bound (ELBO) with the reparameterisation trick. Today it often works behind the scenes — as the image compressor in latent diffusion.
  • Dyskretne kody obrazu i dźwięku (VQ-VAE, VQGAN) (Vector-quantised autoencoders (VQ-VAE, VQGAN)): An autoencoder whose encoder outputs indices of the nearest vectors in a learned codebook, producing discrete tokens for generative models.
  • Rozplątane reprezentacje (β-VAE) (Disentangled representations (β-VAE)): A VAE variant with weight β>1 on the KL term, strengthening independence of latent variables at the cost of reconstruction quality.
• Autoregresyjne generowanie obrazu i dźwięku (Autoregressive image and audio generation): Models generating data as a sequence of elements (pixels, samples, tokens or scales), each conditioned on earlier ones; a rival and complement to diffusion.
  • Generowanie piksel po pikselu i próbka po próbce (PixelCNN, WaveNet) (Pixel- and sample-level autoregressive models (PixelCNN, WaveNet)): Autoregressive convolutional networks modelling pixels or audio samples sequentially using masked or dilated causal convolutions.
  • Maskowane transformery generatywne (MaskGIT) (Masked generative image transformers (MaskGIT)): A bidirectional transformer generating image tokens in parallel over a few iterations, unmasking the most confident predictions from a masked grid.
  • Przewidywanie kolejnej skali (od szkicu do detalu) (Next-scale prediction (visual autoregressive modelling, VAR)): Autoregression over scales: a transformer predicts a whole higher-resolution token map conditioned on all lower ones, instead of tokens in raster order.
• Generatywne sieci przeciwstawne (GAN) (Generative adversarial networks (GAN)): Two networks trained in an adversarial game: a generator creates samples and a discriminator distinguishes them from real data; the discriminator's signal trains the generator to better match the data distribution. Training can be unstable and suffer from mode collapse.
  • Generator sterowany stylem (StyleGAN) (Style-based GAN generators (StyleGAN)): A GAN generator architecture where a mapping network produces a style vector injected into every synthesis layer, separating coarse attributes from fine detail.
  • Przekładanie obrazu na obraz (pix2pix, CycleGAN) (Image-to-image translation (pix2pix, CycleGAN)): Conditional GANs learning mappings between image domains, from paired examples (pix2pix) or without pairs via a cycle-consistency loss (CycleGAN).
  • Stabilne trenowanie GAN (odległość Wassersteina, kara gradientu) (Stable GAN training: Wasserstein loss and gradient penalty): A family of GAN objective changes where a critic estimates the Wasserstein distance between distributions, with the Lipschitz constraint enforced by a gradient penalty.
• Modele dyfuzyjne (Diffusion models): Generating by gradual denoising.
  • Dyfuzja w przestrzeni pikseli (Pixel-space diffusion): Diffusion trained and sampled directly on the pixel grid, without an autoencoder.
    • Klepsydrowy transformer dyfuzyjny (Hourglass diffusion transformer): An hourglass-shaped transformer that scales linearly with pixel count.
    • Przesunięcie harmonogramu szumu (Noise-schedule shifting for high-resolution pixels): An adjustment of log signal-to-noise ratio as the pixel count grows.
    • Transformer pikselowy o dwóch poziomach (Dual-level pixel diffusion transformer): A single-stage diffusion transformer with a patch level and a pixel level.
  • Dyfuzja w przestrzeni ukrytej (Stable Diffusion) (Latent diffusion (Stable Diffusion)): Diffusion performed in an autoencoder's compressed latent space rather than directly in pixel space, reducing computation; text conditioning can be implemented through cross-attention. This design is used by Latent Diffusion models, including Stable Diffusion.
  • Dyfuzyjne modele językowe (Diffusion language models): A family of language models that learn to reverse token masking or noising, rather than only next-token prediction.
    • Dyfuzja blokowa (Block diffusion language models): An autoregressive distribution over blocks, with discrete diffusion inside a block and causal attention across blocks.
    • Dyfuzja ciągła na zanurzeniach słów (Continuous diffusion on word embeddings): Continuous diffusion in embedding space: Gaussian noise on vectors, then rounding back to a token.
    • Dyskretna dyfuzja ze score entropy (Score entropy discrete diffusion): Discrete diffusion trained with score entropy, which matches ratios of token probabilities.
    • Dyskretna dyfuzja ze strukturalną macierzą przejść (Structured discrete diffusion (D3PM)): Discrete diffusion with a transition matrix: uniform, neighbour-based, or with a mask as an absorbing state.
    • Maskowana dyfuzja tekstu (Masked diffusion language models): Diffusion whose only corruption is a mask token: the objective is a weighted mixture of masked-language losses.
      • Próbkowanie z ponownym maskowaniem (Remasking samplers for masked diffusion): Masked-diffusion samplers that can mask previously predicted tokens again and revise them over subsequent steps.
  • Kaskadowe modele dyfuzyjne (Cascaded diffusion models): A chain of diffusion models at increasing resolution, from sketch to detail.
    • Augmentacja obrazu warunkującego (Conditioning augmentation for cascaded diffusion): Corrupting the low-resolution input while training super-resolution models.
  • Sterowanie promptem bez klasyfikatora (classifier-free guidance) (Classifier-free guidance): A technique to strengthen adherence to the prompt: one model learns to generate with and without the condition (dropped randomly in training), and at sampling time the output is pushed toward the conditional version. The guidance scale is the familiar “prompt fidelity vs diversity” knob.
  • Szybkie generowanie w kilku krokach (Fast and few-step diffusion sampling): Methods for shortening diffusion sampling: better solvers without retraining and distillation into few-step and one-step models.
    • Modele spójności (generowanie jednym krokiem) (Consistency models): Models trained to map any point on a denoising trajectory to its endpoint, enabling one- and few-step sampling; trained by distillation or standalone.
    • Przeciwstawna destylacja dyfuzji (generowanie w 1–4 krokach) (Adversarial diffusion distillation): Distilling large diffusion models to 1–4 steps by combining a score-based teacher signal with an adversarial loss.
    • Stopniowa destylacja kroków (Progressive distillation of diffusion samplers): Iterative distillation where a student learns to reproduce two steps of the teacher's deterministic sampler in one, halving the step count each round.
    • Szybsze samplery bez ponownego treningu (DDIM, solvery ODE) (Fast training-free samplers (DDIM, ODE solvers)): Deterministic or less stochastic sampling procedures for trained diffusion models, based on non-Markovian processes and probability-flow ODE solvers.
  • Transformery w modelach dyfuzyjnych i przepływowych (DiT) (Transformers for diffusion and flow models (DiT)): Generative architectures in which a transformer processes patches of an image representation instead of a conventional U-Net backbone. DiT applied this design to latent diffusion, while newer variants also use transformers in flow-trained models.
  • Zaszumianie i odszumianie krok po kroku (DDPM) (Forward noising and learned denoising (DDPM)): The basic construction of denoising diffusion probabilistic models: a forward process gradually adds noise to data, while a model learns to reverse the corruption, often by predicting the added noise. Sampling-step count affects cost and quality, and separate fast-sampling or distillation techniques can reduce it.
• Modele oparte na energii (Energy-based models (EBM)): Models defining an unnormalised density through an energy function, trained via MCMC maximum likelihood, score matching or noise-contrastive estimation.
• Przepływy normalizujące i flow matching (Normalising flows and flow matching): Models that turn simple noise into data with an invertible transform (normalising flows) or a learned velocity field along a path from noise to data (flow matching, rectified flow). Flow matching includes diffusion as a special case and underpins many new image models, e.g. Stable Diffusion 3.
  • Prostowanie przepływu (rectified flow) (Rectified flow): A flow-matching variant learning a velocity field along straight lines between noise and data, with optional repeated straightening (reflow).
  • Przepływy odwracalne z warstwami sprzężenia (RealNVP, Glow) (Coupling-layer normalising flows (RealNVP, Glow)): Generators built from invertible transformations with tractable Jacobian determinants, giving exact log-likelihood via the change-of-variables formula.

Related topics

• Wnioskowanie przybliżone: próbkowanie i metody wariacyjne (Approximate inference: MCMC sampling and variational methods)

Sources

• Stanford CS336 · Language Modeling from Scratch (2025) · reference: Budowa modeli, dane, trening, systemy, inferencja i post-training.
• Stanford CS231n · Computer Vision (2026) · reference: Sieci neuronowe i rozumienie obrazów.
• Stanford CS229 · Machine Learning (2018, archiwum) · reference: Klasyczne ML, statystyka i podstawy uczenia. Użyte jako stabilny szkielet, nie przegląd nowości.
• DeepLearning.AI — course catalogue · course catalogue: 42 topic labels with counts, about 130 courses.