# LongVie
LongVie is an open research framework for controllable ultra-long video generation, released in August 2025 by a consortium of authors from NVIDIA, the [Shanghai AI Laboratory](https://www.edgepedia.org/shanghai-ai-laboratory), Nanjing University, Fudan University, S-Lab, and [Nanyang Technological University](https://www.edgepedia.org/nanyang-technological-university).[1](https://arxiv.org/html/2508.03694v1) It targets a problem that ordinary text-to-video models do not address: keeping characters, scenes, and motion consistent across minutes of generated video rather than seconds, while letting the user steer the result with structural control signals. A second version, LongVie 2, followed in December 2025 with a new backbone and a claimed maximum generation length of five minutes.[2](https://arxiv.org/html/2512.13604)
**What makes it a long-video model.** Standard video diffusion models generate a fixed short clip; extending them by concatenating clips produces drift in appearance, identity, and motion. The LongVie authors attribute this degradation to three causes: separate noise initialization for each clip, control signals normalized independently per clip, and guidance limited to a single modality.[1](https://arxiv.org/html/2508.03694v1) LongVie is autoregressive over clips, with designs intended to carry state across clip boundaries.
| Fact | Value |
|---|---|
| First release | LongVie, arXiv 2508.03694, August 2025[1](https://arxiv.org/html/2508.03694v1) |
| Follow-up | LongVie 2, arXiv 2512.13604, December 2025[2](https://arxiv.org/html/2512.13604) |
| Backbones | CogVideoX (v1); Wan2.1-I2V-14B (v2)[1](https://arxiv.org/html/2508.03694v1) • [2](https://arxiv.org/html/2512.13604) |
| Training data | 130,000 videos (v1); ~100,000 videos (v2)[1](https://arxiv.org/html/2508.03694v1) • [2](https://arxiv.org/html/2512.13604) |
| Demonstrated length | Over one minute (v1); up to five minutes claimed (v2)[1](https://arxiv.org/html/2508.03694v1) • [2](https://arxiv.org/html/2512.13604) |
| Resolution | 480×720 (v1); 352×640 (v2, all experiments)[1](https://arxiv.org/html/2508.03694v1) • [2](https://arxiv.org/html/2512.13604) |
| Inference cost | ~45 min per minute of video on one A100 (v1); ~50 min per video on 8×H100 (v2)[1](https://arxiv.org/html/2508.03694v1) • [3](https://github.com/vchitect/LongVie) |
| Availability | Code on GitHub (Vchitect); LongVie 2 weights on Hugging Face[3](https://github.com/vchitect/LongVie) • [4](https://huggingface.co/Vchitect/LongVie2) |
## Architecture and training as published
**Version 1 (August 2025).** LongVie combines dense and sparse control in a single multimodal framework: dense guidance comes from depth maps, sparse guidance from keypoints, and both are normalized globally rather than per clip. A unified noise initialization keeps the diffusion starting point consistent across successive clips, and degradation-aware training teaches the model to recover from the quality loss that accumulates when each generated clip conditions the next.[1](https://arxiv.org/html/2508.03694v1) The implementation is built on [CogVideoX](https://www.edgepedia.org/cogvideox), realized by copying and fine-tuning 18 DiT blocks (diffusion transformer blocks) in each model.[1](https://arxiv.org/html/2508.03694v1)
The disclosed training set is 130,000 videos drawn from ACID, VchitectT2VDataVerse, and MovieNet, cut into 49-frame clips at 480×720 resolution and 8 fps. Depth maps were produced by Video Depth Anything and sparse tracks by SpatialTracker with 4,900 points per video; captions were generated automatically by Qwen2.5-VL-7B.[1](https://arxiv.org/html/2508.03694v1) Training ran on 8 A100 GPUs for about 3,000 iterations with an effective batch size of 64, taking about five days, using AdamW at learning rate 1e-4.[1](https://arxiv.org/html/2508.03694v1)
**Version 2 (December 2025).** LongVie 2 replaces the backbone with Wan2.1-I2V-14B and builds a dedicated control branch by replicating the first 12 DiT blocks, split along the feature dimension into dense and sparse pathways; the sparse signal is a pointmap rather than keypoints.[2](https://arxiv.org/html/2512.13604) Training proceeds in three stages: multi-modal guidance, degradation-aware training on the input frame, and history-context guidance, which aligns contextual information across adjacent clips.[2](https://arxiv.org/html/2512.13604) Around 100,000 videos were used: about 60,000 from ACID, Vchitect-T2V-DataVerse, and MovieNet in the first two stages (81-frame clips at 352×640, 16 fps), and about 40,000 longer videos from OmniWorld and SpatialVID in the third stage.[2](https://arxiv.org/html/2512.13604) All three stages ran on 16 A100 GPUs with batch size 1 per GPU, for approximately 5,000, 1,000, and 2,000 iterations, about two days in total, at AdamW learning rates of 1e-5 and 5e-6.[2](https://arxiv.org/html/2512.13604)
Beyond generation, version 1 supports long-range video editing and motion or scene transfer, using FLUX's fill and depth-to-image models to synthesize initial frames.[1](https://arxiv.org/html/2508.03694v1)
## Releases and versions
LongVie appeared as arXiv paper 2508.03694 in August 2025, built on CogVideoX.[1](https://arxiv.org/html/2508.03694v1) LongVie 2 followed as arXiv 2512.13604 in December 2025, moving to the Wan2.1-I2V-14B backbone, adopting the three-stage training scheme, and extending the claimed generation length to five minutes.[2](https://arxiv.org/html/2512.13604) The LongVie 2 weights are publicly released on [Hugging Face](https://www.edgepedia.org/hugging-face) under the Vchitect organization.[4](https://huggingface.co/Vchitect/LongVie2) The authors describe a progressive strategy: first enhancing controllability, then extending toward long-term, high-quality generation.[5](https://vchitect.github.io/LongVie2-project/)
One attribution point is disputed between the two papers: the August 2025 paper states that it introduces the LongVGenBench evaluation benchmark, while the December 2025 paper presents LongVGenBench as its own benchmark. Both describe it as 100 high-resolution videos of over one minute each; the record does not resolve the priority claim.[1](https://arxiv.org/html/2508.03694v1) • [2](https://arxiv.org/html/2512.13604)
## By the numbers
All quantitative results below are vendor-reported, meaning the authors' own evaluations in their papers.
**LongVie (v1).** The authors introduce LongVGenBench, a benchmark of 100 high-resolution videos spanning real-world and synthetic environments, each lasting over one minute.[1](https://arxiv.org/html/2508.03694v1) On their comparison table, LongVie scores SSIM 0.557 and LPIPS 0.290 against CogVideoX's 0.374 and 0.521, and leads all listed baselines including DAS and Go-With-Flow, with subject consistency (S.C.) 87.12% and temporal fluency (T.F.) 98.43%.[1](https://arxiv.org/html/2508.03694v1)
**LongVie 2 (v2).** On the paper's comparison table, LongVie 2 scores aesthetic quality (A.Q.) 58.47%, image quality (I.Q.) 69.77%, SSIM 0.529, LPIPS 0.295, subject consistency 91.05%, background consistency 92.45%, overall consistency (O.C.) 23.37%, and dynamic degree (D.D.) 82.95%. Against Wan2.1 the paper lists SSIM 0.406, LPIPS 0.488, A.Q. 49.72%, and D.D. 15.15%; HunyuanGameCraft is listed at D.D. 80.46%. The authors claim state-of-the-art in long-range controllability, temporal coherence, and visual fidelity.[2](https://arxiv.org/html/2512.13604)
**Inference cost.** Version 1 takes about 4.5 minutes to sample a 6-second video, so a one-minute controllable video takes about 45 minutes on a single A100 GPU at 480×720.[1](https://arxiv.org/html/2508.03694v1) The LongVie 2 repository reports roughly 50 minutes per video on 8×H100 GPUs, and a community fork reports a 5-second clip takes about 8 to 9 minutes on a single A100.[3](https://github.com/vchitect/LongVie)
## Licensing, availability and adoption
The official code repository is maintained under the Vchitect organization on GitHub, which describes LongVie 2 as a multimodal controllable world model generating ultra-long videos with depth and pointmap control signals.[3](https://github.com/vchitect/LongVie) LongVie 2 weights are published on Hugging Face under the same organization.[4](https://huggingface.co/Vchitect/LongVie2)
## References
1. [LongVie: Multimodal-Guided Controllable Ultra-Long Video Generation (arXiv, August 2025)](https://arxiv.org/html/2508.03694v1)
2. [LongVie 2: Multimodal Controllable Ultra-Long Video World Model (arXiv, December 2025)](https://arxiv.org/html/2512.13604)
3. [Vchitect/LongVie (GitHub repository)](https://github.com/vchitect/LongVie)
4. [Vchitect/LongVie2 (Hugging Face model card)](https://huggingface.co/Vchitect/LongVie2)
5. [LongVie 2 project page](https://vchitect.github.io/LongVie2-project/)
---
*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › Model families and named models › Video generation models*
*Initially written Sep 17, 2026 · Edited: Oct 11, 2026; Oct 15, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license