Advertise List Your Tool

Switti

Paid

Designing Scale-Wise Transformers for Text-to-Image Synthesis

4.5

Image GeneratorsContact

Inputs: textOutputs: image

Type

Saas

Company

Yandex Research

Visit Website

About Switti

Switti is a scale-wise transformer architecture for text-to-image synthesis, designed by Yandex Research and presented at CVPR 2025. It adapts next-scale prediction autoregressive (AR) architectures to T2I generation while addressing training stability issues. The model introduces a non-causal transformer variant that enables approximately 21% faster sampling and lower memory usage with slightly better generation quality. Additionally, it reveals that classifier-free guidance at high-resolution scales is often unnecessary and can degrade performance; disabling guidance at these scales yields an additional ~32% acceleration and improves fine-grained detail generation. Switti achieves state-of-the-art results among AR models and competes with diffusion models while being up to 7× faster in generating 1024×1024 images.

Key Features

Scale-wise transformer architecture adapted from next-scale prediction AR models

Non-causal transformer design enabling ~21% faster sampling and lower memory usage

Adaptive classifier-free guidance that disables guidance at high-resolution scales for ~32% additional acceleration

Layer-wise calibration procedure for training stability

Up to 7× faster inference compared to state-of-the-art diffusion models at 1024×1024 resolution

Pros & Cons

Pros

Significantly faster generation than diffusion models (up to 7×)
Non-causal architecture improves sampling speed and reduces memory without quality loss
Disabling high-resolution guidance boosts speed and fine-detail quality
Outperforms existing AR models and competes with diffusion models in human evaluations
Open-source code and demo available (GitHub link on page)

Cons