| 文件名 (Filename) | 中文介绍 (Chinese Description) | English Description |
|---|---|---|
NetaAniLumina_Alpha_full_round7_ep4_s68303.pth |
截止至5月22日最新一个完整的epoch，分辨率为1280*1280。 | The latest complete epoch as of May 22nd, with a resolution of 1280*1280. |
NetaAniLumina_Alpha_full_round5_ep5_s182000.pth |
截止至5月8日，此为分辨率提升前的最后一个检查点(ckpt)，分辨率为1024*1024。 | As of May 8th, this was the last checkpoint (ckpt) before the resolution increase, with a resolution of 1024*1024. |
模型介绍：neta-art/NetaAniLumina_Alpha
欢迎体验 neta-art/NetaAniLumina_Alpha！
这是一款由 Nieta.art 团队基于优秀的开源模型 Lumina-Image-2.0 by Alpha-VLLM 进一步训练的二次元文本生成图像（T2I）模型。在训练过程中，我们利用了大量丰富、多样化的二次元图像数据集，并为其配备了高质量的自然语言处理（NLP）标签，力求模型能够更好地理解和诠释二次元相关的文本描述。
开放测试与共享精神：
本着技术共享与开源社区共同进步的精神，我们决定在模型训练的极早期阶段（Alpha测试）即公开部分预训练权重。我们希望通过这种方式，能够让感兴趣的开发者和研究者们提前接触到这款模型，进行初步的探索、测试，并期待未来能获得宝贵的反馈。
⚠️ 重要注意事项与免责声明 ⚠️
在您下载和使用本模型前，请务必仔细阅读并理解以下几点：
- 极早期阶段： 本模型目前仍处于非常非常初期的开发与训练阶段。这意味着模型远未成熟，可能存在大量已知和未知的问题、缺陷以及不稳定的行为。
- 无后续优化： 当前发布的Alpha版本没有进行任何后续的美学对齐、风格强化、指令跟随优化或其他任何形式的后训练（post-training）阶段。模型的原始输出可能与您的期望有较大差距。
- 使用难度极高： 由于上述原因，当前模型的使用门槛非常高，可能需要特定的参数调整和大量的尝试才能生成特定内容，且结果可能不尽人意。请有充分的心理准备。
- 关于分辨率提升与知识遗忘： 模型近期进行了一次分辨率提升的训练调整，目前正处于重新适应阶段。这可能会导致短期内出现较多的知识遗忘现象，影响生成效果的稳定性。
- 必要组件与使用资源： 此页面表格中仅发布DiT（Diffusion Transformer）部分的权重 (请直接点击文件名下载)。模型的其他必要组件（如VAE、文本编码器等）请访问 Comfy-Org/Lumina_Image_2.0_Repackaged 页面获取。为帮助您快速上手，我们还在根目录提供了：
- 入门手册 (简版):
guide.md - ComfyUI 工作流:
lumina_workflow.json
- 入门手册 (简版):
- 请勿分享模型权重： 为确保项目在开发阶段的可控性，请勿在任何公开或私下渠道再次分发或共享此Alpha版本的模型权重文件。
- 无授权声明： 至关重要的一点是，本次Alpha测试版本的发布，不构成对模型本身、模型权重、以及模型所产生内容的任何形式的使用授权或许可。 由于模型完全没有训练完毕，其性质和能力仍在探索中。Nieta.art 保留对模型及其衍生物的所有权利。请勿将此早期测试模型及其产出物用于任何商业用途、正式项目或任何可能产生不良影响的场景。
我们感谢您对 Nieta.art 和 NietaAniLumina 项目的关注与兴趣。我们期待在模型成熟后，能为大家带来更完善、更强大的版本。
Model Introduction: neta-art/NetaAniLumina_Alpha
Welcome to neta-art/NetaAniLumina_Alpha!
This is an anime-style text-to-image (T2I) model further trained by the Nieta.art team, based on the excellent open-source model Lumina-Image-2.0 by Alpha-VLLM. During the training process, we utilized an extensive, rich, and diverse dataset of anime images, paired with high-quality Natural Language Processing (NLP) tags, striving for the model to better understand and interpret anime-related text descriptions.
Open Testing & Spirit of Sharing:
In the spirit of technological sharing and the mutual progress of the open-source community, we have decided to release partial pre-trained weights during the extremely early stages of model training (Alpha test). We hope that this will allow interested developers and researchers to get an early touchpoint with this model for preliminary exploration and testing, and we look forward to receiving valuable feedback in the future.
⚠️ IMPORTANT NOTES AND DISCLAIMERS ⚠️
Before you download and use this model, please carefully read and understand the following points:
- Extremely Early Stage: This model is currently in a very, very initial phase of development and training. This means the model is far from mature and may contain numerous known and unknown issues, defects, and unstable behaviors.
- No Post-Optimization: The currently released Alpha version has NOT undergone any subsequent aesthetic alignment, style enhancement, instruction following optimization, or any other form of post-training. The raw output of the model may significantly differ from your expectations.
- Extremely High Difficulty of Use: Due to the reasons above, the barrier to entry for using the current model is very high. It may require specific parameter tuning and extensive experimentation to generate specific content, and the results might not be satisfactory. Please be fully prepared for this.
- Regarding Resolution Increase & Knowledge Forgetting: The model has recently undergone a training adjustment for a resolution increase and is currently in a re-adaptation phase. This may lead to more significant temporary knowledge forgetting in the short term, affecting the stability of generation results.
- Required Components & Usage Resources: This page's table only releases the DiT (Diffusion Transformer) weights (click filenames directly to download). Other necessary components for the model (such as VAE, text encoders, etc.) can be found on the Comfy-Org/Lumina_Image_2.0_Repackaged page. To help you get started, we also provide the following in the root directory:
- Quick Start Guide (Simplified):
guide.md - ComfyUI Workflow:
lumina_workflow.json
- Quick Start Guide (Simplified):
- Do Not Share Model Weights: To ensure the controllability of the project during its development phase, please do not redistribute or share the model weight files of this Alpha version through any public or private channels.
- No Authorization Granted: Crucially, the release of this Alpha test version does NOT constitute any form of usage authorization or license for the model itself, its weights, or the content generated by it. As the model is entirely unfinished in its training, its nature and capabilities are still under exploration. Nieta.art reserves all rights to the model and its derivatives. Please do not use this early test model or its outputs for any commercial purposes, official projects, or any scenarios that could lead to adverse effects.
We thank you for your interest in Nieta.art and the NietaAniLumina project. We look forward to bringing you a more complete and powerful version once the model matures.