🌌 Wan2.2-Anime-Unlimited (v1.0 Stable Toy)

The First Truly Open-Source, Unrestricted Anime Video Model

首个真正开源、无限制的二次元视频生成模型

Based on Wan2.2-5B | Runs on 6GB VRAM | Uncensored 基于 Wan2.2-5B 架构 | 6GB 显存即可运行 | 无限制内容

📢 Project Mission / 计划愿景

I am building the ultimate open-source anime video generation model. 我正在打造最完美、唯一的开源二次元视频大模型。

My goal is to create a 5B model that outperforms the 27B base model in the specific domain of anime aesthetics and battle scenes. 我的目标不仅仅是做一个玩具，而是打造一个在二次元美学和战斗场景领域，表现超越 27B 基座模型的一流模型。

🚧 Current Status: v1.0 Stable (Qualified Toy)

This is the v1.0 "Qualified Toy" version. 这是 v1.0 “合格的玩具” 版本。

Compared to the Alpha version, v1.0 is fully stable. It serves as a solid proof of concept, demonstrating that a 5B model, with 531 high-quality polished clips, can generate impressive anime content. 相比 Alpha 版本，v1.0 是完全稳定的正式版。作为一个“合格的玩具”，它完美证明了 5B 模型在使用 531 个高质量精修视频微调后，也能产出令人惊艳的二次元内容。

Training Data: 531 High-Quality Clips (Polished & Optimized).
Performance: Stable & Validated.
Specialty: Character interactions & Basic movements.

🚀 Roadmap: The Path to Supremacy (路线图)

We are moving fast. Here is the plan for the next major updates: 我们正在快速迭代。以下是接下来的宏大计划：

✅ v1.0 (Current): "The Qualified Toy" (合格的玩具)

Data: 531 clips.
Goal: Stability test passed. A fun and usable model for everyone.
Status: Released.

🛠️ v2.x Series: "Battle & Life" (战斗与日常)

Data: 4,000+ High-Quality Videos.
Focus: Intense Battle Scenes (Sakuga) & Slice of Life.
New Feature: Native support for 8s & 10s generation! (Time duration optimization).
Target: In the vertical anime domain, surpass the performance of Wan2.2-27B.
Plan: Releasing in sub-versions (v2.1, v2.2...) to refine details.
目标： 在二次元垂直领域，效果力压 27B 版本！并将全面支持 8秒与 10秒的长视频生成优化。

🌟 v3.0: "Concept Expansion" (概念爆发)

Data: 10,000+ Videos.
Focus: Massive expansion of anime concepts, art styles, and complex interactions.
Goal: The ultimate 5B anime model.

🔊 v5.0 - v6.0: "Multimodal Evolution" (多模态进化)

I will introduce a new Audio Model between v5 and v6! Yes, she will no longer be an old-fashioned silent video model, but a multimodal audio-video model with sound alignment and unrestricted creativity! This is set in stone, because I have already prepared the code. I am just waiting to finish training the silent models before conducting a secondary training on the audio video models. The only thing that might defeat me is money?

同时我将在v5到v6版本之间加入新的音频模型！ 是的，她将不再是老套的无声视频模型，而是拥有声音对齐，且无任何限制可自由创造的多模态声音视频模型！这是板上钉钉的事，因为我目前已经处理好代码，只待将所有没有声音视频模型训练完毕后，再将有声音的视频模型统一进行二次训练，能够击败我的或许只有金钱？

📚 Example Prompts (Demo Pack)

Please download the Wan2.2_Demo_Pack.zip from the Files tab! 请务必下载文件列表中的 Wan2.2_Demo_Pack.zip！ (Follow the examples to get the best result / 请参照示例以获得最佳效果)

💰 Support the Revolution / 支持这项革命

Training on 4,000+ videos for v2.x requires massive GPU power. 为了训练 v2 版本的 4000+ 视频集，我们需要燃烧大量的显卡经费。

If you want to see v2.0 (The 27B Killer) sooner, please support me! 如果你想尽快玩到“力压 27B”的 v2.0 版本，请支持我！

👇 Click below to donate / 点击下方赞助 👇

🌍 Global: Ko-fi

🇨🇳 China: Afdian / 爱发电

Powered by the Open Source Community & Wan2.2

💬 Join the Community / 加入交流讨论

QQ Group: 838904299

Downloads last month: 961