[ComfyUI]最强腾讯开源混元视频炼丹炉已就绪,国漫经典李慕婉,一致性写真视频轻松批量直出

科技   2025-01-04 07:55   浙江  

混元视频炼丹炉,一致性视频LoRA

🌹大家好!欢迎来到破狼公众号。感谢大家的支持与鼓励。在AIGC探索道路上,我将与你一路同行。喜欢就星标关注破狼公众号或文末扫码加入交流群 !

diffusion-pipe简介

近期文章中多次介绍到最新腾讯混元视频模型。同时混元视频社区已支持视频LORA的训练,这也是不仅是一致性和可空性解决方案,也是视频题材数据集丰富的重要方式之一。当前相对成熟的视频LORA训练方案是diffusion-pipe。这是一个可以用于扩散模型的并行训练脚本。目前支持 FluxLTX-Video 和 HunyuanVideo。当然目前社区使用它最多的还是因为腾讯混元视频LORA训练

当前diffusion-pipe还不支持Window部署安装,对于Window用户建议使用WSL 2部署,同时建议新开启Conda等虚拟环境部署,独立隔离环境测试。关于详细的部署方案,请参见部署文档:https://civitai.com/articles/9798/training-a-lora-for-hunyuan-video-on-windows

在部署时候需要注意,需要使用 @kijai大佬发布模型 配置对应训练配置LLM参数,可以与ComfyUI工作流共享路径。当前训练大约需要24G显存,可以基于图片或者切分视频数据集训练。需要根据数据集数量重复次数周期数设置对应配置参数,这需要来自个人训练经验,欢迎大家更多心得交流。

  • • Github:https://github.com/tdrussell/diffusion-pipe
  • • HunyuanVideo_comfy:https://huggingface.co/Kijai/HunyuanVideo_comfy/tree/main
  • • llava-llama-3-8b-text-encoder-tokenizer:https://huggingface.co/Kijai/llava-llama-3-8b-text-encoder-tokenizer/tree/main
  • • clip-vit-large-patch14:https://huggingface.co/openai/clip-vit-large-patch14/tree/main

针对本地显存不足同学,可以使用笔者在云端部署的镜像训练。具体训练指南请参加对应镜像操作指南文档。

  • • 仙宫云混元视频LORA训练丹炉:https://www.yuque.com/yuqueyonghuwhsczq/isrggf/xg7cw0ic1irct3go?singleDoc#
  • • 晨羽智云混元视频LORA训练丹炉:https://www.yuque.com/yuqueyonghuwhsczq/isrggf/edb1qdbfe8ugft7u?singleDoc

混元视频LORA ComfyUI体验

下面采用LIBLIBAI平台最新训练的混元视频LORA体验,具体本地LORA使用工作流部署参见之前的文章:[ComfyUI]腾讯混元视频:开源LORA生态来啦!视频人物一致性&运镜可控性里程碑,支持NSFW

混元视频LORA可LIBLIBAI平台获取

  • • 混元视频LORA:https://www.liblib.art/modelinfo/2f4a59b8eaf643d580f421c8c78f3d93?versionUuid=3c17059e7b994165a47abedbd3524e7f
  • • 在线混元视频工作流:https://www.liblib.art/modelinfo/2f4a59b8eaf643d580f421c8c78f3d93?versionUuid=3c17059e7b994165a47abedbd3524e7f

01.舞蹈

pl, LiMuwan, The video is a portrait of a young Asian LiMuwan. Her hair is dark brown and she has a tiara on her head. She is wearing a light purple dress with a sweetheart neckline. The dress is strapless and she is wearing long, dangling earrings. Over her dress is a light pink fur coat that is draped over her shoulders. There is a shadow casted on the wall behind the LiMuwan., LiMuwan moves his body in a slow dance

02.皇冠

pl, LiMuwan, side vciew face, A LiMuwan is wearing a red robe with gold embroidery on it. She has a gold crown on top of her head. There is a red lantern hanging behind her., LiMuwan Walking back and forth as if waiting for someone

03.独自哀伤

pl, LiMuwan, A LiMuwan is wearing a white dress. There is a blue belt around her waist. The LiMuwan has long dark hair and a crown on top of her head. There are flowers coming off of a tree behind her., LiMuwan Looking up at the sky, the flowers were blown to the ground by the wind, crying independently

混元视频模型:关注公众号口令【混元视频模型获取

更多推荐文章:

• [ComfyUI]腾讯混元视频:官方极限优化8GB可运行!32G到8G极限优化,开源生态加速

• [ComfyUI]腾讯混元视频:强烈推荐!质量增强神器!Enhance-A-Video无性能和内存显著影响

• [ComfyUI]HelloMemeV2:表情迁移神器!直播数字人,更强保真和表情丰富的一致性迁移

• 腾讯混元视频:ComfyUI官方原生支持!GGUF量化低显存福音!开源生态高速发展

感兴趣加入[AGI技术交流群]+V

    如果觉得文章不错,就请在看转发三连

破狼
关注AIGC、LLM、绘图作品、软件工程、技术学习。交流+V:shunshizhiwu。
 最新文章