Organizations: William & Mary · Carnegie Mellon University · University of Illinois Urbana-Champaign · Squirrel Ai Learning · University of Wisconsin–Madison
Unified multimodal models combine image understanding and generation within one system, creating an opportunity to improve both capabilities through paired image--text supervision. We examine the outputs of a pretrained unified model in both directions and observe a loss of specificity: generated images can contain the requested objects with incorrect attributes, while captions identify a scene without expressing its distinguishing details. These failures call for supervision that specifies both the visual content to realize and the descriptive differences to distinguish. We propose UniMore, a post-training method that derives these complementary learning tasks from the fine-grained content of image--text pairs. The generation route predicts grounded crops from their corresponding phrases, matching the scope of the condition and visual target while using the native visual objective. The understanding route combines positive-caption fitting with ranking against captions that omit or alter details, conditioned on the full image. The two routes are optimized independently, and their learned updates are combined into a single unified model without changing its inference procedure. Experiments show gains across multiple understanding and generation benchmarks.
Figures & tables
Backbone
G/U
LR
Steps
Warm.
Procs
BAGEL
G
5×10−5
1,000
50
4
U
1.8×10−5
600
60
2
Janus-Pro
G
1×10−5
250
25
1
U
1.8×10−5
600
60
1
Show-o2
G
5×10−6
500
50
1
U
1.8×10−5
300
30
1
Table 1: Training settings. G/U denote generation/understanding.
MathVista
Method
Size
MMMU
MMBench
MM-Vet
MMStar
MC
FF
All
Avg.
Native unified multimodal models
BAGEL ( 2025 )
7B
51.90
82.65
65.90
63.20
80.19
61.52
71.60
67.05
Janus-Pro ( 2025b )
7B
40.70
67.64
33.70
42.00
51.30
32.83
42.80
45.37
Janus ( 2025 )
1.3B
27.30
27.69
27.00
26.67
35.37
16.30
26.60
27.05
JanusFlow ( 2025 )
1.3B
29.00
64.16
31.80
39.19
39.07
29.78
34.80
39.79
Table 2: Multimodal understanding. Bold and underline mark the best and second-best results within the BAGEL post-training group.
WISE
Method
UEval
Culture
Time
Space
Bio.
Phys.
Chem.
All
Native unified multimodal models
BAGEL
30.90
38.83
43.86
47.14
36.20
42.05
29.40
39.89
Janus-Pro
20.60
36.16
38.53
47.89
36.05
47.45
24.85
38.11
Janus
–
20.80
27.07
35.08
17.05
19.10
10.95
22.22
JanusFlow
–
27.31
32.22
39.47
32.15
28.60
19.05
29.64
Table 3: Generation performance. Bold and underline mark the best and second-best results within the BAGEL post-training group.
RealUnify
Unison
Method
GEU
UEG
IC
UGG
GGU
ME
Mean
BAGEL
39.00
22.00
80.51
72.68
32.65
35.41
55.31
UniMore
40.00
23.67
79.57
71.33
33.20
57.40
60.37
Table 4: Cross-task consistency and synergy on BAGEL.
Understanding
Generation
Variant
MMMU
MMStar
MathVista
WISE
UEval
BAGEL
51.90
63.20
71.60
39.89
30.90
G-only
53.00
68.73
73.10
43.52
31.33
U-only
53.00
70.40
73.20
39.67
30.48
UniMore
53.30
70.13
73.80
43.48
31.60
Table 6: Ablation of the two training routes on BAGEL. Bold and underline mark the best and second-best available results.
Supervision
WISE
BAGEL
39.89
Caption → Image
40.18
Phrase → Image
42.00
Phrase → Crop
43.52
UniMore
43.48
Table 7: Supervision-target comparisons on BAGEL. (a) Visual prediction targets. (b) Caption supervision, where “+” adds ranking to positive-caption CE.
Unified multimodal models are envisioned to bridge the gap between understanding and generation. Yet, to achieve competitive performance, state-of-the-art models adopt largely decoupled understanding and generation components. This design, while effective for individual tasks, weakens the connection required for mutual enhancement, leaving the potential synergy empirically uncertain. We propose to explicitly restore this synergy by introducing Understanding-Oriented Post-Training (UNO), a lightweight framework that treats understanding not only as a distinct task, but also a direct supervisory signal to steer generative representations. By incorporating objectives that encode semantic abstraction (captioning) and structural details (visual regression), we enable effective gradient flow from understanding to generation. Extensive experiments on image generation and editing demonstrate that understanding can serve as an effective catalyst for generation.
Zeyu Liu, Zanlin Ni, Yang Yue +5
1Tsinghua University · 2Kolors Team, Kuaishou Technology
Recently, unified multimodal models (UMMs) have made remarkable progress in integrating visual understanding and generation, demonstrating strong potential for complex text-to-image (T2I) tasks. Despite their theoretical promise, a persistent capability gap exists: UMMs typically exhibit superior visual understanding but comparatively weaker generative capabilities. This discrepancy arises largely from the intrinsic decoupling between the understanding and generation processes. While a UMM can accurately interpret fine-grained visual details, it often struggles to produce semantically coherent images from complex textual prompts. To address this challenge, we explore UMMs' internal understanding capability to enhance generation quality. We propose a token-level intrinsic text-image alignment reward mechanism, GvU, enabling the UMM to act simultaneously as teacher and student: it evaluates its own outputs using the understanding branch to guide the generations accordingly. Building upon this, we design a self-supervised reinforcement learning framework, allowing UMMs to iteratively improve their generation quality through understanding-based intrinsic reward signals--without reliance on external supervision. Experimental results show that our method substantially boosts UMMs' generation, which in turn strengthens their fine-grained visual understanding, narrowing the capability gap between UMMs' visual understanding and generation.
Jiadong Pan, Liang Li, Yuxin Peng +6
Institute of Computing Technology, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Peking University +2
While unified multimodal models (UMMs) jointly perform visual understanding and generation within a single model, functional unification does not guarantee learning synergy: the two objectives may reinforce each other, compete for capacity, or merely coexist. We investigate their relationship at the representation, task, and system levels in a controlled, structurally native setting without pretrained vision priors. At the representation level, we find that each objective provides useful signal to the other: generation enriches the visual features learned for understanding, while understanding strengthens vision--language alignment for generation. However, when both objectives are forced through the same computation path, one tends to dominate. A task-decoupled architecture that specializes conflicting visual computation while preserving semantic interaction avoids this asymmetric degradation. At the task level, through three case studies, we find positive bidirectional transfer when understanding and generation tasks rely on shared knowledge. At the system level, we show that an end-to-end UMM outperforms a matched planner--executor pipeline on complex tasks that explicitly require both image understanding and generation. Together, these results show that the value of UMMs extends beyond a unified interface: appropriate specialization, shared task knowledge, and end-to-end optimization can turn coexistence into synergy.
Penghao Wu, Haiwen Diao, Weichen Fan +3
1S-Lab, Nanyang Technological University · 2SenseTime Research