Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Generation

TL;DR

Tuna-2 introduces a unified multimodal model that directly uses pixel embeddings for understanding and generation, eliminating the need for pretrained vision encoders and achieving state-of-the-art results.

Contribution

It presents a novel encoder-free approach that simplifies architecture and enhances multimodal understanding and generation directly from raw pixels.

Findings

Tuna-2 achieves state-of-the-art performance on multimodal benchmarks.

Encoder-free design converges slower but yields stronger understanding at scale.

Pretrained vision encoders are not necessary for high-quality multimodal modelling.

Abstract

Unified multimodal models typically rely on pretrained vision encoders and use separate visual representations for understanding and generation, creating misalignment between the two tasks and preventing fully end-to-end optimization from raw pixels. We introduce Tuna-2, a native unified multimodal model that performs visual understanding and generation directly based on pixel embeddings. Tuna-2 drastically simplifies the model architecture by employing simple patch embedding layers to encode visual input, completely discarding the modular vision encoder designs such as the VAE or the representation encoder. Experiments show that Tuna-2 achieves state-of-the-art performance in multimodal benchmarks, demonstrating that unified pixel-space modelling can fully compete with latent-space approaches for high-quality image generation. Moreover, while the encoder-based variant converges faster in early pretraining, Tuna-2's encoder-free design achieves stronger multimodal understanding at scale, particularly on tasks requiring fine-grained visual perception. These results show that pretrained vision encoders are not necessary for multimodal modelling, and end-to-end pixel-space learning offers a scalable path toward stronger visual representations for both generation and perception.