image synthesis / generative models
Official code for training VQGAN and autoregressive transformers for high-resolution image synthesis.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Choose for autoregressive high-resolution image generation with bitwise tokens instead of VQGAN+AR.
[CVPR 2025 Oral]Infinity ∞ : Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis
Choose for text-to-image high-resolution synthesis with a linear diffusion transformer instead of VQGAN+AR.
SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer
Choose for patch-based diffusion high-resolution generation if you do not want a VQGAN+transformer pipeline.
Let us democratise high-resolution generation! (CVPR 2024)
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Choose for older multimodal text/image/video synthesis if you need a unified transformer and can accept an archived codebase.
A unified 3D Transformer Pipeline for visual synthesis
Choose for layout-conditioned image synthesis with an older GAN if you do not need a transformer.
Semantic Image Synthesis with SPADE
Choose for semantic-layout-to-photo cascaded refinement synthesis if you need an older, narrower conditional model.
Photographic Image Synthesis with Cascaded Refinement Networks
These projects were analysed and named taming-transformers among their alternatives. The relationship is not symmetric — how taming-transformers rates them is a separate judgement, made when taming-transformers is analysed in its own right.
calls taming-transformers “Choose when you want the older VQGAN+Transformer image-synthesis research codebase; no modern text-to-image/video pipeline.”
SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer