Generative Image Modelling Methods (GANs, Diffusion and Image Translation)
Core methods for learning to generate or translate images, from GANs and autoencoders to latent diffusion.
Tier Grades
Tier A: Frontier AI labs
feasible
Frontier labs build on these methods at scale.
Tier B: Indonesian AI labs
feasible
Training GANs and small diffusion models fits an Indonesian lab.
Tier C: Institution units such as ITB
feasible
Small academic teams wrote most of these methods, such as the 5-author latent diffusion paper.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Delivery teams use finished models rather than new methods; method work is a stretch.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataNot in the sources read; latent diffusion runs diffusion in the latent space of a pretrained autoencoder.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 5 authors on the latent diffusion paper (arXiv 2112.10752).
- Time horizonYears.
- Shelf-life signalsFlow matching replaces diffusion training, Autoregressive image tokens compete with diffusion
- Pipeline roleTier C and academia invent methods; Tier A scales them; Tier D uses the resulting models, for example for synthetic training data.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Latent Diffusion (LDM)image generationexemplar
DDPM (Denoising Diffusion Probabilistic Models)image generationexemplar
StyleGAN2image generationexemplarCycleGANimage generationexemplar
STARFlowimage generationproduced by
DDColorimage generationproduced byEG3D3Dproduced byStyleGAN3image generationproduced by
Guided Diffusion (ADM)image generationproduced byImproved DDPMimage generationproduced by
InstColorizationimage generationproduced by
CoCosNetimage generationproduced by
BigBiGANimage generationproduced by
SinGANimage generationproduced byStyleGANimage generationproduced byBigGAN (incl. BigGAN-Deep)image generationproduced byvid2vidvideo generationproduced by
SAGAN (Self-Attention GAN)image generationproduced bySPIRALimage generationproduced bypix2pixHDimage generationproduced by
StarGAN (v1, v2)image generationproduced by
SPGANimage generationproduced byVQ-VAEimage generationproduced byProgressive GAN (PGGAN)image generationproduced by+7
Arbitrary Image Stylization (Magenta)image generationproduced bypix2piximage generationproduced byColorful Image Colorizationimage generationproduced by
Fast Neural Style Transferimage generationproduced by
DCGANimage generationproduced by
XiNet Picasso Muse style transferimage generationproduced byStyleGANimage generationproduced byLinks
R&D TopicGenerative Image Modelling Methods (GANs, Diffusion and Image Translation)
ModelsStyleGAN2NVIDIA, image generationCycleGANUC Berkeley, image generationDDPM (Denoising Diffusion Probabilistic Models)UC Berkeley, image generationLatent Diffusion (LDM)LMU Munich, image generation+27 more
StyleGAN2NVIDIA, image generationCycleGANUC Berkeley, image generationDDPM (Denoising Diffusion Probabilistic Models)UC Berkeley, image generationLatent Diffusion (LDM)LMU Munich, image generation+27 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 31
StyleGANNVIDIA, image generationproduced byLatent Diffusion (LDM)LMU Munich, Runway, image generationexemplarSPGANUniversity of Chinese Academy of Sciences, University of Technology Sydney, image generationproduced byBigBiGANGoogle DeepMind, image generationproduced byCoCosNetUniversity of Science and Technology of China (USTC), Microsoft Research, image generationproduced byColorful Image ColorizationUC Berkeley, image generationproduced byDCGANindico, Meta, image generationproduced byDDColorAlibaba DAMO, image generationproduced byEG3DNVIDIA, 3Dproduced byInstColorizationNational Tsing Hua University, image generationproduced bySAGAN (Self-Attention GAN)Rutgers University, Google, image generationproduced bySinGANTechnion - Israel Institute of Technology, Google, image generationproduced by+19
STARFlowApple, image generationproduced byStarGAN (v1, v2)Korea University, Naver, image generationproduced byStyleGANNVIDIA, image generationproduced byStyleGAN2NVIDIA, image generationexemplarStyleGAN3NVIDIA, image generationproduced byvid2vidNVIDIA, video generationproduced byXiNet Picasso Muse style transferSTMicroelectronics, image generationproduced byArbitrary Image Stylization (Magenta)Google, image generationproduced byBigGAN (incl. BigGAN-Deep)Google DeepMind, image generationproduced byCycleGANUC Berkeley, image generationexemplarDDPM (Denoising Diffusion Probabilistic Models)UC Berkeley, image generationexemplarFast Neural Style TransferStanford University, image generationproduced byGuided Diffusion (ADM)OpenAI, image generationproduced byImproved DDPMOpenAI, image generationproduced bypix2pixUC Berkeley, image generationproduced bypix2pixHDNVIDIA, image generationproduced byProgressive GAN (PGGAN)NVIDIA, image generationproduced bySPIRALGoogle DeepMind, image generationproduced byVQ-VAEGoogle DeepMind, image generationproduced byPapers
- Denoising Diffusion Probabilistic ModelsJonathan Ho et al., arXiv preprint, 2020 founding
- Score-Based Generative Modeling through Stochastic Differential EquationsYang Song et al., ICLR 2021 (per arXiv comment), 2020 founding
- Elucidating the Design Space of Diffusion-Based Generative ModelsTero Karras et al., NeurIPS 2022 (per arXiv comment), 2022 state of the art
- Scalable Diffusion Models with TransformersWilliam Peebles et al., arXiv preprint, 2022 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2112.10752 https://arxiv.org/abs/2112.10752
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Generative Image Modelling Methods (GANs, Diffusion and Image Translation)", https://aurion.semai.id/topics/generative-image-modelling-methods/, records of October 2026.