Lance: Unified Multimodal Modeling by Multi-Task Synergy

Fengyi Fu*, Mengqi Huang*†, Shaojin Wu*, Yunsheng Jiang*, Yufei Huo, Jianzhu Guo†§
Hao Li, Yinghang Song, Fei Ding, Qian He, Zheren Fu, Zhendong Mao, Yongdong Zhang
* Equal contribution † Corresponding authors § Project lead
Intelligent Creation Team, ByteDance

Lance is a lightweight native unified multimodal model for image and video understanding, generation, and editing, trained from scratch within a training budget of no more than 128 GPUs using a staged multi-task recipe.

Paper Code Hugging Face

Text-to-Video

Nine text-conditioned cases focused on character motion, fantasy animals, two-person interaction, and cinematic dreamlike scenes.

Surfing animal character
Robot and dog duet
Wizard cat with magic
Tropical sunset shoreline
Motorcycle through canyon
Pottery character motion
Piano performance
Robot boxing match
Two-person embrace

Video Editing

Nine prompt-driven single- and multi-function editing cases spanning background transformation, object addition and removal, subject replacement, appearance restyling, stylization, and action edits.

Background transformation
Object addition
Character replacement
Subject replacement
Stylized scene
Multi-function winter edit
Action and expression
Multi-function edit
Object removal

Multi-turn Consistency Editing

Source video followed by four linked edits on the same subject: replacement, accessory addition, background rewrite, and motion update.

Source video
Hair replacement
Floral headband
Castle background
Hand wave

Intelligent Video Generation

Structured planning and physics-oriented examples that probe control over multi-step spatial behavior.

4x4 grid navigation, route 1
5x5 grid navigation, route 1
6x6 grid navigation, route 1
4x4 grid navigation, route 2
5x5 grid navigation, route 2
6x6 grid navigation, route 2

Video Understanding

Selected video question answering and captioning cases that evaluate temporal reasoning, motion recognition, and concise-to-detailed description.

VQA question

How many times did the person launch objects on the table?
Options:
(A) 3
(B) 2
(C) 4

Answer

(A) 3

VQA question

The person makes sets of repeated actions. How many distinct repeated actions did the person do?
Options:
(A) 2
(B) 3
(C) 4

Answer

(A) 2

VQA question

In which direction does the purple sphere move in the video?
Options:
(A) Down and to the right.
(B) Up and to the left.
(C) Up and to the right.
(D) The object is stationary.

Answer

(A) Down and to the right.

VQA question

What is the unrealistic phenomenon displayed in the video?
Options:
(A) The man can manipulate time via phone.
(B) Man grabs an object through a phone screen.
(C) Chocolate transforms into different objects.
(D) Visible means of propulsion enables flight.

Answer

(B) Man grabs an object through a phone screen.

Short caption

Offer a succinct account of the culinary process shown in this video.

Answer

Add tomato puree and mix it well with chicken pieces.

Long caption

Provide a detailed description of the given video, capturing its key moments.

Answer

In a sunlit meadow, a small tortoiseshell butterfly rests on a purple flower. A bee, with black and yellow stripes, lands on the same flower. The butterfly flaps its wings gently, while the bee busies itself, collecting nectar. The flower sways slightly in the breeze. The butterfly then takes off, and the bee follows, both heading to the next flower. The scene is a vivid display of insect interaction in a natural setting, with the colors of the butterfly and the bee contrasting against the green background of the meadow. The video captures this peaceful moment in a short 6-second duration.

Text-to-Image

Representative text-to-image outputs spanning photorealistic, stylized, compositional, and typography-heavy prompts.

Text-to-Image

Image Editing

Instruction-guided image editing cases showing local replacement, style transfer, object-aware modifications, and layout-preserving transformations.

Image editing examples

Image Understanding

Six selected visual question answering cases spanning charts, trade data, OCR, documents, landmarks, and natural phenomena.

Pie chart used for segment comparison reasoning

Is the largest segment greater than sum of all the other segments?

Answer Yes, the largest segment in the pie chart is the blue one, which is the most populous segment.

Export partner chart for Honduras trade question

Which country was Honduras' most important export partner in 2019?

Answer El Salvador was the most important exporter in 2019, as he had a significant share in the United States.

Car image used for license plate recognition

What is the license plate number of the car?

Answer The license plate number of the car is BX62 BFY.

Market research chart for promotional meeting and event spending

According to the data from the proprietary market research,how much amount was spent on the promotional meetings and events during 1998?

Answer According to the data from the proprietary market research, the total amount spent on the promotional meetings and events during 1998 was approximately $1.3 billion.

Colosseum image for appearance description

What is the appearance of the Colosseum in Rome, Italy?

Answer The Colosseum in Rome, Italy is a magnificent ancient amphitheater made of stone and concrete. It is characterized by its large, arched openings and its semi-circular shape. The Colosseum has been partially restored, and its original color has faded, but the architectural details, like the columns and the semi-circular arches, are still visible. The lighting in the image highlights the grandeur and historical significance of the Colosseum, making it stand out against the backdrop of the evening sky.

Total solar eclipse image for visual knowledge answering

How does a total solar eclipse look like from Earth?

Answer A total solar eclipse looks like a dark circle with a bright, white outer edge. This is a result of the Earth's rotation and the Earth's shadow, which partially obscures the sun's light. The bright white edge of the eclipse is caused by the sun's high energy and its interaction with the Earth's atmosphere, while the dark part of the eclipse is due to the Earth's shadow and the surrounding air currents. The solar eclipse's shape, with its bright white edge and dark center, is similar to the shape of a full moon or a dark disk. It is a natural phenomenon that occurs in the atmosphere of the Earth and is an important part of the solar system.

Framework

Lance keeps a shared interleaved sequence for text, image, and video context, then separates semantic understanding and visual synthesis through dedicated expert pathways.

Lance Framework
The framework combines semantic ViT tokens for understanding, clean/noisy VAE latents for generation, generalized 3D causal attention, and MaPE to reduce positional interference among heterogeneous visual tokens.

Comparison on multimodal benchmarks

Radar charts compare Lance against BAGEL, UniWorld-V1, Wan2.1-T2V, Hunyuan Video, Open-Sora-2.0, Gen-3, and other baselines across image, editing, and video benchmarks.

Radar chart comparing Lance with image generation, editing, and video generation baselines

Image generation on GenEVAL

GenEVAL measures object count, color, position, and attribute binding. Lance ties the best overall score among listed unified models while remaining a compact 3B model.

Scroll horizontally to inspect all metrics.

GenEVAL image generation results. Higher scores are better.
Method # Params. Overall Single Obj. Two Obj. Counting Colors Position Color Attri.
Generation-only models
FLUX.1-dev 12B 0.82 0.98 0.93 0.75 0.93 0.68 0.65
GPT Image 1 - 0.84 0.99 0.92 0.85 0.92 0.75 0.61
Qwen-Image 20B 0.87 0.99 0.92 0.89 0.88 0.76 0.77
Unified models
MetaQuery-XL† 7B 0.80 - - - - - -
OmniGen2 4B 0.80 1.00 0.95 0.64 0.88 0.55 0.76
Show-o2 7B 0.76 1.00 0.87 0.58 0.92 0.52 0.62
UniWorld-V1 13B 0.80 0.99 0.93 0.79 0.89 0.49 0.70
BAGEL† 7B 0.88 0.98 0.95 0.84 0.95 0.78 0.77
Mogao 7B 0.89 1.00 0.97 0.83 0.93 0.84 0.80
TUNA 7B 0.90 1.00 0.97 0.81 0.91 0.88 0.83
Lance 3B 0.90 1.00 0.94 0.84 0.97 0.87 0.81

Image generation on DPG-Bench

DPG-Bench stresses complex prompt following across global, entity, attribute, relation, and other compositional dimensions; Lance is especially strong on relation grounding.

Scroll horizontally to inspect all metrics.

DPG-Bench image generation results. Higher scores are better.
Method # Params. Overall Global Entity Attribute Relation Other
Generation-only models
PixArt-a 0.6B 71.11 74.97 79.32 78.60 82.57 76.96
SDXL 3.5B 74.65 83.27 82.43 80.91 86.76 80.41
Hunyuan-DiT 1.5B 78.87 84.59 80.59 88.01 74.36 86.41
Playground v2.5 - 75.47 83.06 82.59 81.20 84.08 83.50
DALL-E 3 - 83.50 90.97 89.61 88.39 90.58 89.83
SD3-Medium 2B 84.08 87.90 91.01 88.83 80.70 88.68
Emu3-Gen 8B 80.60 85.21 86.68 86.84 90.22 83.15
FLUX.1-dev 12B 83.84 74.35 90.00 88.96 90.87 88.33
Qwen-Image 20B 88.32 91.32 91.56 92.02 94.31 92.73
Unified models
Emu3-DPO 8B 81.60 - - - - -
Janus - 79.68 82.33 87.38 87.70 85.46 86.41
Janus-Pro-7B 7B 84.19 86.90 88.90 89.40 89.32 89.48
Ovis-U1 1.2B 83.72 82.37 90.08 88.68 93.35 85.20
OmniGen2 4B 83.57 88.81 88.83 90.18 89.37 90.27
Show-o2 7B 86.14 89.00 91.78 89.96 91.81 91.64
UniWorld-V1 13B 81.38 83.64 88.39 88.44 89.27 87.22
BAGEL† 7B 85.07 88.94 90.37 91.29 90.82 88.67
Mogao 7B 84.33 82.37 90.03 88.26 93.18 85.40
InternVL-U 1.7B 85.18 90.39 90.78 90.68 90.29 88.77
TUNA 7B 86.76 90.42 91.68 90.94 91.87 90.73
Lance 3B 84.67 83.89 91.07 89.36 93.38 80.80

Image editing on GEdit-Bench

GEdit-Bench evaluates instruction-guided edits such as background, color, material, subject, style, and tone changes. Lance reports the best average score among listed unified models.

Scroll horizontally to inspect all metrics.

GEdit-Bench image editing results. Higher scores are better.
Method # Params. Avg/G-O BC CA MM MC PB ST SA SR SRp TM TT
Generation-only models
Gemini 2.0 - 6.32 - - - - - - - - - - -
GPT Image 1 - 7.49 6.96 6.85 7.10 5.41 6.74 7.44 7.51 8.73 8.55 8.45 8.69
Qwen-Image-Edit 20B 8.01 8.23 8.30 7.33 8.05 7.49 6.74 8.57 8.09 8.29 8.48 8.50
Unified models
Lumina-DiMOO 8B 3.91 3.43 4.27 3.08 2.77 4.74 5.19 4.44 3.80 4.38 2.68 4.20
Ovis-U1 1.2B 6.42 7.49 6.88 6.21 4.79 5.98 6.46 7.49 7.25 7.27 4.48 6.31
BAGEL 7B 6.52 7.32 6.91 6.38 4.75 4.57 6.15 7.90 7.16 7.02 7.32 6.22
InternVL-U 1.7B 6.66 7.08 7.05 6.38 7.02 6.03 6.27 7.13 6.55 6.33 6.59 6.85
InternVL-U (w/ CoT) 1.7B 6.88 7.05 7.87 6.50 6.99 5.77 6.10 7.33 7.16 7.12 7.36 6.46
Lance 3B 7.30 7.73 7.74 7.28 7.83 7.50 7.03 7.64 7.85 7.71 4.46 7.57

Video generation on VBench

VBench covers video quality, semantic alignment, object attributes, spatial relations, and motion-related dimensions. Lance obtains the top total score in the unified model group.

Scroll horizontally to inspect all metrics.

VBench video generation results. Higher scores are better.
Model # Params. Total Score↑ Quality Score Semantic Score Subj. Consist. Bkg. Consist. Temp. Flicker Motion Smooth. Dynamic Degree Aesthetic Quality Imaging Quality Object Class Multi. Objects Human Action Color Spatial Relation Scene Appear. Style Temp. Style Overall Consist.
Generation-only models
ModelScope 1.7B 75.75 78.05 66.54 89.87 95.29 98.28 95.79 66.39 52.06 58.57 82.25 38.98 92.40 81.72 33.68 39.26 23.39 25.37 25.67
LaVie 3B 77.08 78.78 70.31 91.41 97.47 98.30 96.38 49.72 54.94 61.90 91.82 33.32 96.80 86.39 34.09 52.69 23.56 25.93 26.41
Show-1 6B 78.93 80.42 72.98 95.53 98.02 99.12 98.24 44.44 57.35 58.66 93.07 45.47 95.60 86.35 53.50 47.03 23.06 25.28 27.46
AnimateDiff-V2 - 80.27 82.90 69.75 95.30 97.68 98.75 97.76 40.83 67.16 70.10 90.90 36.88 92.60 87.47 34.60 50.19 22.42 26.03 27.04
VideoCrafter-2.0 - 80.44 82.20 73.42 96.85 98.22 98.41 97.73 42.50 63.13 67.22 92.55 40.66 95.00 92.92 35.86 55.29 25.13 25.84 28.23
CogVideoX 5B 81.61 82.75 77.04 96.23 96.52 98.66 96.92 70.97 61.98 62.90 85.23 62.11 99.40 82.81 66.35 53.20 24.91 25.38 27.59
Kling - 81.85 83.39 75.68 98.33 97.60 99.30 99.40 46.94 61.21 65.62 87.24 68.05 93.40 89.90 73.03 50.86 19.62 24.17 26.42
Open-Sora-2.0 - 81.71 82.10 80.14 98.75 98.00 99.40 99.49 20.74 64.33 65.62 94.50 77.72 95.40 85.98 76.18 52.71 22.98 25.91 27.57
Gen-3 - 82.32 84.11 75.17 97.10 96.62 98.61 99.23 60.14 63.34 66.82 87.81 53.64 96.40 80.90 65.09 54.57 24.31 24.71 26.69
Step-Video-T2V 30B 81.83 84.46 71.28 98.05 97.67 99.40 99.08 53.06 61.23 70.63 80.56 50.55 94.00 88.25 71.47 24.38 23.17 26.01 27.12
Hunyuan Video - 83.43 85.07 76.88 97.22 97.60 99.39 99.05 71.94 60.28 67.24 83.48 66.71 94.40 89.79 72.13 54.46 22.21 24.52 26.95
Wan2.1-T2V 14B 83.69 85.59 76.11 97.52 98.09 99.46 98.30 65.46 66.07 69.43 86.28 69.58 95.40 88.59 75.39 45.75 22.64 23.19 25.91
Unified models
HaploOmni 7B 78.10 - - 96.40 97.60 - 96.80 65.30 - - - - - - - 34.60 - - -
Emu3 8B 80.96 - - 95.32 97.69 - 98.93 79.27 59.64 - 86.17 44.64 77.71 - 68.73 37.11 20.92 - -
VILA-U 7B 74.01 76.26 65.04 - - - - - - - - - - - - - - - -
Show-o2 2B 81.34 82.10 78.31 97.28 96.78 97.68 98.25 40.83 65.15 67.06 94.81 76.01 95.20 80.89 62.61 57.67 23.29 25.27 27.00
TUNA 1.5B 84.06 84.32 83.04 95.99 96.72 98.02 98.33 69.39 65.88 66.83 95.41 92.31 97.50 87.67 78.12 58.59 23.18 24.68 27.71
Lance 3B 85.11 85.14 84.96 94.52 94.28 99.66 95.93 75.83 64.33 66.78 96.58 93.86 97.80 92.61 93.61 64.75 23.14 25.53 27.04

Video understanding on MVBench

MVBench evaluates video understanding across action, object, spatial, temporal, and reasoning categories. Lance achieves the best average score among listed unified models.

Scroll horizontally to inspect all metrics.

MVBench video understanding results. Higher scores are better.
Model # Params. Avg.↑ AS AP AA FA UA OE OI OS MD AL ST AC MC MA SC CO EN ER CI
Understanding-only models
Video-LLaMA 7B 34.1 27.5 25.5 51.0 29.0 39.0 48.0 40.5 38.0 22.5 22.5 43.0 34.0 22.5 32.5 45.5 40.0 30.0 21.0 37.0
LLaMA-Adapter 7B 31.7 23.0 28.0 51.0 30.0 33.0 53.5 32.5 33.5 25.5 21.5 30.5 29.0 22.5 41.5 39.5 31.5 22.5 28.0 32.0
Video-ChatGPT 7B 32.7 23.5 26.0 62.0 22.5 26.5 54.0 28.0 40.0 23.0 20.0 31.0 30.5 25.5 39.5 48.5 33.0 29.5 26.0 35.5
VideoChat 7B 35.5 33.5 26.5 56.0 33.5 40.5 53.0 40.5 30.0 25.5 27.0 48.5 35.0 20.5 42.5 46.0 41.0 23.5 23.5 36.0
VideoChat2 7B 51.1 66.0 47.5 83.5 49.5 60.0 58.0 71.5 42.5 23.0 23.0 88.5 39.0 42.0 58.5 44.0 36.5 35.0 40.5 65.5
ST-LLM 7B 54.9 66.0 53.5 84.0 44.0 58.5 80.5 73.5 38.5 42.5 31.0 86.5 36.5 56.5 78.5 43.0 46.5 34.5 41.5 58.5
GPT-4V - 43.5 55.5 63.5 72.0 46.5 73.5 18.5 59.0 29.5 12.0 40.5 83.5 39.0 12.0 22.5 45.0 52.0 31.0 59.0 11.0
PLLaVA 34B 58.1 67.5 53.0 82.0 47.0 79.0 68.5 67.5 36.5 37.5 49.5 91.0 40.5 43.0 70.0 51.5 66.5 39.5 63.5 59.0
Video-CCAM 9B 64.6 83.0 67.0 89.5 49.0 72.0 86.5 81.0 45.0 28.0 29.0 90.0 59.0 67.0 85.0 63.5 77.0 34.0 73.5 59.0
Qwen2.5-VL 3B 67.0 - - - - - - - - - - - - - - - - - - -
TimeMarker 8B 67.4 79.0 74.5 89.0 53.5 77.0 94.0 76.0 41.5 52.5 47.0 91.5 53.0 76.5 92.5 57.0 70.5 23.5 53.5 82.5
InternVideo2 7B 67.3 86.0 70.0 87.0 56.0 75.0 91.0 86.0 40.0 48.0 53.0 90.0 41.0 73.0 92.0 52.0 56.0 33.0 57.0 74.0
Unified models
Show-o2 1.5B 50.6 63.8 59.5 63.5 40.0 70.5 54.5 66.0 36.5 36.0 27.0 88.0 43.5 43.0 58.0 44.5 54.0 28.5 39.5 45.0
Show-o2 7B 55.7 60.1 67.0 68.0 45.5 78.0 51.0 73.5 44.5 36.0 39.0 92.5 51.5 36.0 59.5 52.0 64.0 38.0 60.0 43.0
TUNA 1.5B 54.4 - - - - - - - - - - - - - - - - - - -
UniVideo 7B 46.3 54.3 41.5 77.5 50.0 62.5 68.2 50.5 37.5 36.0 29.5 35.5 28.5 52.5 70.5 33.5 40.5 37.5 36.5 38.0
Lance 3B 62.0 73.9 76.5 71.5 49.0 63.5 96.0 72.5 33.0 63.5 33.0 86.0 41.0 82.0 97.5 43.0 47.5 31.5 40.0 77.0

Citation

@misc{lance2026,
  title  = {Lance: Unified Multimodal Modeling by Multi-Task Synergy},
  author = {Fengyi Fu and Mengqi Huang and Shaojin Wu and Yunsheng Jiang and Yufei Huo and Jianzhu Guo and Hao Li and Yinghang Song and Fei Ding and Qian He and Zheren Fu and Zhendong Mao and Yongdong Zhang},
  year   = {2026},
  note   = {Manuscript}
}
Demo

Video preview