Research map / Vision-language and multimodal / Human and embodied modeling
TUNA: Taming Unified Visual Representations for Native Unified Multimodal Models
Zhiheng Liu, Weiming Ren, Haozhe Liu, Zijian Zhou, Shoufa Chen, Haonan Qiu, Xiaoke Huang, Zhaochong An, Fanny Yang, Aditya Patel, Viktar Atliha, Tony Ng and 13 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use unified VAE+SigLIP2 representation instead of decoupled SigLIP2/VAE representations (ablation) | TUNA 1.5B ablation · understanding and generation benchmarks | SEED / GenEval: SEED 61.4 -> 66.5, GenEval 78.3 -> 79.4 | Improved |
| use TUNA unified representation instead of Show-o2 late-fusion (ablation) | TUNA 1.5B ablation · Und & Gen training | GenEval: 75.9 -> 79.4 | Improved |
| use SigLIP 2 instead of SigLIP as representation encoder (ablation) | TUNA 1.5B ablation · Und & Gen training | SEED / GenEval: SEED 64.6 -> 66.5, GenEval 76.9 -> 79.4 | Improved |
| use DINOv3 instead of SigLIP 2 as encoder (ablation) | TUNA 1.5B ablation · Und & Gen training | GenEval: 79.4 -> 78.9 (MME-p 1361 -> 1396) | No clear effect |
| add generation data to understanding training (ablation) | TUNA (SigLIP 2) 1.5B · understanding benchmarks | MMMU / SEED / GQA: MMMU 37.6 -> 38.1, SEED 62.9 -> 66.5 | Improved |
| add understanding data to generation training (ablation) | TUNA (SigLIP 2) 1.5B · generation benchmarks | GenEval: 77.8 -> 79.4 | Improved |
| add generation data to understanding training on MME-p (ablation) | TUNA (SigLIP 2) 1.5B · MME perception | score: 1386 -> 1361 | Hurt |
Later papers that compared with this method
4 later accepted papers report Tuna as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Generation (NeurIPS 2026 · better on 3 of 6 shared results)
- Advancing Analytic Class-Incremental Learning through Vision-Language Calibration (ICML 2026 · better on 4 of 4 shared results)
- STARFlow2: Bridging Language Models and Normalizing Flows for Unified Multimodal Generation (NeurIPS 2026 · better on 0 of 2 shared results)
- UNICOM: Unified Multimodal Modeling via Compressed Continuous Semantic Representations (NeurIPS 2026 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Show-o2 — Show-o2: Improved Native Unified Multimodal Models (NeurIPS 2025) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Generation (NeurIPS 2026 · similarity 0.93)
- Generation Enhances Understanding in Unified Multimodal Models via Multi-Representation Generation (ICML 2026 · similarity 0.91)
- Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation (CVPR 2025 · similarity 0.90)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.89)
- UNICOM: Unified Multimodal Modeling via Compressed Continuous Semantic Representations (NeurIPS 2026 · similarity 0.89)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
