Add license, base model, and full author list to model card

#1
by nielsr HF Staff - opened
Files changed (1) hide show
  1. README.md +30 -60
README.md CHANGED
@@ -1,4 +1,9 @@
1
  ---
 
 
 
 
 
2
  tags:
3
  - audio-visual
4
  - world-model
@@ -6,9 +11,6 @@ tags:
6
  - real-time
7
  - streaming-generation
8
  - video-generation
9
- language:
10
- - en
11
- pipeline_tag: any-to-any
12
  ---
13
 
14
  # MaineCoon: Pursuing A Real-Time Audio-Visual Social World Model
@@ -17,91 +19,59 @@ pipeline_tag: any-to-any
17
 
18
  | | |
19
  |---|---|
20
- | 🌐 Project | https://mainecoon.tech/ |
21
- | 🕹️ Experience | https://mainecoon.tech/experience-platform |
22
- | 📄 Paper (arXiv) | https://arxiv.org/abs/2606.17800 |
23
- | 📝 Blog | https://mainecoon.tech/blogs |
24
- | 💻 GitHub | https://github.com/catnip-ai-tech/MaineCoon |
25
 
26
  ## Abstract
27
 
28
- As an increasing majority of global video content is consumed on social platforms for interactive social purposes, video generation models built for social worlds are important but largely overlooked by previous studies. In this work, we define the position of social world models and build a prototype model as the first step towards this goal. While previous world models successfully simulate physical environments or gaming world exploration, they remain fundamentally detached from human-centric social dynamics. They typically omit critical auditory information or fail to capture the high-engagement pacing, emotional resonance, and rapid conversational flow that define viral social media. To bridge this gap as the first step to social world models, we present **MaineCoon**, the first real-time audio-visual autoregressive model that has **22B parameters** and is capable of real-time streaming generation and sub-second interaction, with a record-breaking frame rate of **up to 47.5 FPS, on a single GPU**. To the best of our knowledge, MaineCoon is also the first real-time audio-visual generation model specifically optimized for social-interactive applications. To enable efficient and stable training, we introduce several novel techniques into MaineCoon, including self-resampling, cross-modal representation alignment, domain-aware preference optimization, and reinforced online-policy distillation (ROPD). We also design the first agentic streaming inference framework that supports thousand-second-scale or even longer generation while mitigating drift with agentic cache management and prompt planning. These innovations significantly accelerate training while optimizing real-time inference performance. We believe this work not only sets a new state-of-the-art (SOTA) performance benchmark for high-quality, low-latency, and long-horizon audio-visual autoregressive models, but also points out the paradigm shift desired for next-generation AI-native social platforms.
29
-
30
- ## Highlights
31
 
32
- - **⚡ Real-time on a single GPU.** A 22B interactive audio-visual autoregressive model capable of streaming generation and sub-second interaction, with a record-breaking frame rate of **up to 47.5 FPS** on a single H100. Generation cost drops well **below \$0.001 per second** and keeps falling.
33
- - **🌍 A new paradigm: social world models.** MaineCoon positions and serves as the first generative core for *social world models*, a technical foundation for next-generation AI-native social platforms.
34
- - **🎓 Forcing-free streaming training.** A multi-stage training paradigm — **self-resampling**, **cross-modal representation alignment**, **domain-aware preference optimization**, and **reinforced online-policy distillation (ROPD)** — that enables native, efficient streaming audio-visual training at 22B scale.
35
- - **🧠 Agentic streaming inference.** An agentic inference framework that supports **thousand-second-scale** generation while mitigating drift through agentic cache management, chunk commitment, long-context rollout, and prompt planning.
36
- - **📊 SocialVideo-Bench.** A new benchmark focused on audio-visual social-video generation, with 9 representative metrics covering visual quality, motion, audio quality, audio-visual alignment, and social-video harmony. MaineCoon outperforms 7 representative open audio-visual models while achieving the fastest generation speed — a new state of the art for real-time social video generation.
37
 
38
- ## Showcase
39
-
40
- Hand-picked MaineCoon generations (audio-visual, with sound) play directly in the **[GitHub repository](https://github.com/catnip-ai-tech/MaineCoon)**.
41
 
42
- 🎬 **Minute-scale, long-form demos** are best viewed on our **[blog](https://mainecoon.tech/blogs)**.   🕹️ **Try MaineCoon live** at the **[experience platform](https://mainecoon.tech/experience-platform)**.
 
 
 
 
43
 
44
  ## Benchmark — SocialVideo-Bench
45
 
46
- **Table 2. Main quantitative results on SocialVideo-Bench.** 🐱 **MaineCoon (Ours)** achieves the best average score and wins most metrics, including the two most comprehensive ones — Audio-Visual Harmony (AVH) and Joint Audio-Visual Integrated Score (JAVIS) — over both streaming and bidirectional baselines.
47
 
48
- | Type | Model | Vis↑ | Mot↑ | Aud↑ | IB-TV↑ | IB-TA↑ | IB-AV↑ | AV-Al↑ | AVH↑ | JAVIS↑ | Average↑ |
49
- |:--|:--|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|
50
- | Bidirectional T2AV | JavisDiT++ | 4.39 | **2.22** | 4.06 | 0.134 | 0.070 | 0.151 | 0.312 | 0.136 | 0.112 | 0.711 |
51
- | | Ovi | 4.44 | 1.89 | 3.76 | _0.138_ | 0.079 | 0.191 | **0.412** | 0.188 | 0.162 | 0.779 |
52
- | | JoyAI-Echo | 4.61 | 1.17 | 3.47 | **0.147** | 0.088 | 0.226 | 0.319 | 0.196 | 0.173 | 0.749 |
53
- | | MoVA | _4.66_ | 1.68 | 3.69 | 0.133 | 0.105 | 0.258 | _0.359_ | 0.245 | 0.216 | 0.842 |
54
- | | LTX-2.3 | 4.10 | 0.99 | 4.06 | 0.132 | 0.111 | 0.311 | 0.334 | 0.287 | _0.247_ | 0.848 |
55
- | Streaming TA2V | LiveAvatar | 4.60 | 1.46 | _4.13_ | 0.131 | 0.120 | _0.316_ | 0.326 | _0.291_ | 0.246 | 0.892 |
56
- | | SoulX-FlashTalk | 4.65 | _1.99_ | 4.07 | 0.128 | _0.120_ | 0.307 | 0.279 | 0.283 | 0.238 | _0.895_ |
57
- | **Streaming T2AV** | 🐱 **MaineCoon (Ours)** | **4.71** | 1.62 | **4.35** | 0.127 | **0.130** | **0.318** | 0.334 | **0.308** | **0.272** | **0.934** 🥇 |
58
 
59
- <sub>🐱 = our method &nbsp;·&nbsp; **bold** = best, _italic_ = second best. &nbsp; Metrics Vis: visual quality · Mot: motion · Aud: audio quality · IB-TV / IB-TA / IB-AV: ImageBind Text–Video / Text–Audio / Audio–Video alignment · AV-Al: audio–visual alignment · AVH: Audio-Visual Harmony · JAVIS: Joint Audio-Visual Integrated Score. See the technical report for the full benchmark and metric definitions.</sub>
60
-
61
- **Table 3. Latency and model size comparison.** Sampling throughput (FPS) is measured for 480P 20-second generation on a single H100 GPU. 🐱 **MaineCoon (Ours)** has the **largest model yet by far the fastest** speed — up to **7× faster** than other streaming audio-visual generators, and faster even than a 1.3B streaming video model.
62
 
63
  | Type | Model | Params | FPS↑ |
64
  |:--|:--|:--:|:--:|
65
- | Bidirectional T2AV | JavisDiT++ | 1.8B | 0.87 |
66
- | | Ovi | 11B | 0.58 |
67
- | | JoyAI-Echo | 23B | 18.0 |
68
- | | MoVA | 32B | 0.26 |
69
- | | LTX-2.3 | 22B | 1.40 |
70
- | | LTX-2.3-Distilled | 22B | _20.7_ |
71
  | Streaming T2V | Causal-Forcing | 1.3B | 19.1 |
72
- | | Helios-Distilled | 14B | 18.2 |
73
- | | Krea | 14B | 6.1 |
74
- | Streaming TA2V | LiveAvatar | 14B | 6.7 |
75
- | | SoulX-FlashTalk | 14B | 6.6 |
76
  | **Streaming T2AV** | 🐱&nbsp;**MaineCoon&nbsp;(Ours)** | **22B** | **47.5**&nbsp;🥇 |
77
 
78
- <sub>🐱 = our method &nbsp;·&nbsp; **bold** = best, _italic_ = second best. FPS for 480P-20s on a single H100.</sub>
79
-
80
- ## Paper
81
-
82
- The full paper is available on **[arXiv:2606.17800](https://arxiv.org/abs/2606.17800)**. A PDF copy is also included in this repository: [`MaineCoon_Technical_Report.pdf`](./MaineCoon_Technical_Report.pdf). It covers the social-video data infrastructure, the native streaming autoregressive training recipe, the agentic streaming inference framework, SocialVideo-Bench, and a position/outlook on social world models.
83
-
84
  ## Acknowledgements
85
 
86
  MaineCoon stands on the shoulders of the open-source community. We are especially grateful to:
87
 
88
- - **🎬 LTX-2.3 & the LTX series — [Lightricks](https://github.com/Lightricks).** MaineCoon's audio-visual backbone builds on the excellent open **LTX-2.3** model. Huge credit to the LTX team and the broader LTX-Video series.
89
- - **LTX-2** (incl. LTX-2.3): https://github.com/Lightricks/LTX-2
90
- - **LTX-Video**: https://github.com/Lightricks/LTX-Video
91
- - **⚡ DMD series & the distribution-matching distillation community.** Our reinforced online-policy distillation (ROPD) builds on the **Distribution Matching Distillation (DMD / DMD2)** line of work and the wider few-step / real-time distillation community.
92
- - **DMD2**: https://github.com/tianweiy/DMD2
93
- - **DMD** (project page): https://tianweiy.github.io/dmd/
94
-
95
- We thank these projects and their communities for advancing real-time, few-step, and streaming video generation.
96
 
97
  ## Citation
98
 
99
  ```bibtex
100
  @article{catnip2026mainecoon,
101
  title = {MaineCoon: Pursuing A Real-Time Audio-Visual Social World Model},
102
- author = {Catnip AI Team},
103
  year = {2026},
104
  journal = {arXiv preprint arXiv:2606.17800},
105
  url = {https://arxiv.org/abs/2606.17800}
106
  }
107
- ```
 
1
  ---
2
+ language:
3
+ - en
4
+ license: apache-2.0
5
+ pipeline_tag: any-to-any
6
+ base_model: Lightricks/LTX-Video
7
  tags:
8
  - audio-visual
9
  - world-model
 
11
  - real-time
12
  - streaming-generation
13
  - video-generation
 
 
 
14
  ---
15
 
16
  # MaineCoon: Pursuing A Real-Time Audio-Visual Social World Model
 
19
 
20
  | | |
21
  |---|---|
22
+ | 🌐 Project | [mainecoon.tech](https://mainecoon.tech/) |
23
+ | 🕹️ Experience | [Try it live](https://mainecoon.tech/experience-platform) |
24
+ | 📄 Paper (arXiv) | [arXiv:2606.17800](https://arxiv.org/abs/2606.17800) |
25
+ | 📝 Blog | [Technical Blog](https://mainecoon.tech/blogs) |
26
+ | 💻 GitHub | [catnip-ai-tech/MaineCoon](https://github.com/catnip-ai-tech/MaineCoon) |
27
 
28
  ## Abstract
29
 
30
+ As video content is increasingly consumed on social platforms, video generation models built for social worlds are important but largely overlooked. We present **MaineCoon**, the first real-time audio-visual autoregressive model. With **22B parameters**, it is capable of real-time streaming generation and sub-second interaction, achieving a record-breaking frame rate of **up to 47.5 FPS on a single GPU**.
 
 
31
 
32
+ MaineCoon is optimized for social-interactive applications using several novel techniques: self-resampling, cross-modal representation alignment, domain-aware preference optimization, and reinforced online-policy distillation (ROPD). We also introduce an agentic streaming inference framework that supports thousand-second-scale generation while mitigating drift.
 
 
 
 
33
 
34
+ ## Highlights
 
 
35
 
36
+ - ** Real-time on a single GPU.** Capable of streaming generation and sub-second interaction on a single H100.
37
+ - **🌍 A new paradigm: social world models.** Serves as the first generative core for social world models, a foundation for next-generation AI-native social platforms.
38
+ - **🎓 Forcing-free streaming training.** Multi-stage training enabling native, efficient streaming audio-visual training at 22B scale.
39
+ - **🧠 Agentic streaming inference.** Supports long-horizon generation through agentic cache management and prompt planning.
40
+ - **📊 SocialVideo-Bench.** A new benchmark where MaineCoon outperforms 7 representative open audio-visual models while achieving the fastest generation speed.
41
 
42
  ## Benchmark — SocialVideo-Bench
43
 
44
+ **Main quantitative results on SocialVideo-Bench.** 🐱 **MaineCoon (Ours)** achieves the best average score and wins most metrics, including Audio-Visual Harmony (AVH) and Joint Audio-Visual Integrated Score (JAVIS).
45
 
46
+ | Type | Model | Vis↑ | Mot↑ | Aud↑ | AV-Al↑ | AVH↑ | JAVIS↑ | Average↑ |
47
+ |:--|:--|:--:|:--:|:--:|:--:|:--:|:--:|:--:|
48
+ | Bidirectional T2AV | LTX-2.3 | 4.10 | 0.99 | 4.06 | 0.334 | 0.287 | 0.247 | 0.848 |
49
+ | Streaming TA2V | SoulX-FlashTalk | 4.65 | 1.99 | 4.07 | 0.279 | 0.283 | 0.238 | 0.895 |
50
+ | **Streaming T2AV** | 🐱&nbsp;**MaineCoon&nbsp;(Ours)** | **4.71** | 1.62 | **4.35** | 0.334 | **0.308** | **0.272** | **0.934**&nbsp;🥇 |
 
 
 
 
 
51
 
52
+ **Latency and model size comparison.** Sampling throughput (FPS) measured for 480P 20-second generation on a single H100 GPU.
 
 
53
 
54
  | Type | Model | Params | FPS↑ |
55
  |:--|:--|:--:|:--:|
56
+ | Bidirectional T2AV | LTX-2.3-Distilled | 22B | 20.7 |
 
 
 
 
 
57
  | Streaming T2V | Causal-Forcing | 1.3B | 19.1 |
 
 
 
 
58
  | **Streaming T2AV** | 🐱&nbsp;**MaineCoon&nbsp;(Ours)** | **22B** | **47.5**&nbsp;🥇 |
59
 
 
 
 
 
 
 
60
  ## Acknowledgements
61
 
62
  MaineCoon stands on the shoulders of the open-source community. We are especially grateful to:
63
 
64
+ - **🎬 LTX-2.3 & the LTX series — [Lightricks](https://github.com/Lightricks).** MaineCoon's audio-visual backbone builds on the excellent open **LTX-2.3** model.
65
+ - **⚡ DMD series.** Our reinforced online-policy distillation (ROPD) builds on the **Distribution Matching Distillation (DMD / DMD2)** line of work.
 
 
 
 
 
 
66
 
67
  ## Citation
68
 
69
  ```bibtex
70
  @article{catnip2026mainecoon,
71
  title = {MaineCoon: Pursuing A Real-Time Audio-Visual Social World Model},
72
+ author = {Lichen Bai and Tianhao Zhang and Shitong Shao and Dingwei Tan and Qiyu Zhong and Zhengpeng Xie and Haopeng Li and Qinghao Huang and Dandan Shen and Tengjiao Ji and Wei Wang and Peicheng Wu and Yuxuan Zhao and Xiangyu Zhu and Welly Luo and Shurui Yang and Zeke Xie},
73
  year = {2026},
74
  journal = {arXiv preprint arXiv:2606.17800},
75
  url = {https://arxiv.org/abs/2606.17800}
76
  }
77
+ ```