Скачать или смотреть TUNA: Taming Unified Visual Representations for Native Unified Multimodal Models (Dec 2025)

Title: TUNA: Taming Unified Visual Representations for Native Unified Multimodal Models (Dec 2025)
Link: http://arxiv.org/abs/2512.02014v1
Date: December 2025

Summary:
This paper introduces Tuna, a native Unified Multimodal Model (UMM) that employs a unified continuous visual representation for both understanding and generation tasks. By cascading a VAE encoder with a representation encoder (SigLIP 2), Tuna creates a shared space that supports end-to-end processing for images and videos, eliminating the format mismatches found in decoupled models. The model utilizes an LLM decoder to perform autoregressive text generation and flow matching for visual generation, achieving state-of-the-art performance across diverse benchmarks including image/video understanding, generation, and editing.

Key Topics:
Unified Multimodal Models
Visual Representation Learning
Multimodal Understanding
Image and Video Generation
Flow Matching
Autoregressive Models
Image Editing

Chapters:
00:00 - Introduction to Tuna
01:50 - Visual Encoding Challenges
03:48 - Anchoring in VAE Latents
05:09 - Cascaded Visual Encoder Design
06:38 - Efficient Video Processing
07:16 - Three-Stage Training Protocol
08:58 - Performance and Benchmarks
09:47 - Proving Mutual Enhancement
10:48 - Early vs Late Fusion
12:28 - Qualitative Generation Results
14:19 - Summary and Future Outlook

Stock video credits:
@svetjekolem - https://www.pexels.com/@svetjekolem
Kelly - https://www.pexels.com/@kelly
Bedrijfsfilmspecialist.nl - https://www.pexels.com/@bedrijfsfilms...
Colors Motion Graphics - https://www.pexels.com/@colors-motion...
Trippy Lagoon - https://www.pexels.com/@trippy-lagoon...
Soumya - https://www.pexels.com/@soumya-1446957
Anete Lusina - https://www.pexels.com/@anete-lusina
Pressmaster - https://www.pexels.com/@pressmaster
Pixabay - https://www.pexels.com/@pixabay
Yaroslav Shuraev - https://www.pexels.com/@yaroslav-shuraev
Colin Jones - https://www.pexels.com/@larchmedia
StefWithAnF - https://www.pexels.com/@stefwithanf-1...
tunnel motions - https://www.pexels.com/@tunnelmotions
José Alfredo Munguía Lira - https://www.pexels.com/@rectorretro
Mikhail Nilov - https://www.pexels.com/@mikhail-nilov
crazy motions - https://www.pexels.com/@crazy-motions...
Charlie Mounsey - https://www.pexels.com/@charlie-mouns...
Pachon in Motion - https://www.pexels.com/@pachon-in-mot...
Pavel Danilyuk - https://www.pexels.com/@pavel-danilyuk
Ron Lach - https://www.pexels.com/@ron-lach
cottonbro studio - https://www.pexels.com/@cottonbro
Kindel Media - https://www.pexels.com/@kindelmedia
Danil Shostak - https://www.pexels.com/@danil-shostak...
KATRIN BOLOVTSOVA - https://www.pexels.com/@ekaterina-bol...
Engin Akyurt - https://www.pexels.com/@enginakyurt
Dan Cristian Pădureț - https://www.pexels.com/@paduret
Anthony 🙂 - https://www.pexels.com/@inspiredimages
Silviu Din - https://www.pexels.com/@silviu-din-16...
Stas Knop - https://www.pexels.com/@stasknop