If you like my work and want to support me in a diferent way you can buy me a Ko-Fi if you want.
Updated 2026.06.25: testV2 is retrained based on anima_base V1.0 using 14,000 images. In addition to realistic photos, the dataset also includes @3dcg and Blender style images. The turbo version adds an 8-step LoRa model to the 00006ep .
Turbo Lora is no longer required separately.
Cfg: 1-3, steps 8-24.
2026.06.25更新:testV2基于anima_base V1.0使用14,000张图片重新训练。数据集中除去写实照片之外,还额外有@3dcg和blender风格。turbo版在00006ep基础上添加了8step lora模型。现在不需要单独加turbo lora。Cfg: 1-3, steps 8-24.
喜欢的大佬可以打赏赞助一下。
Updated 2026.06.02: Based on "Anima base1.0", I made some full-scale fine-tuning attempts. When used with the LoRa model I released, the results are even better.
The dataset contains approximately 4,000 to 5,000 images.
All tests were performed using "anima-turbo-lora-v0.2", "8steps", "er_sde", and "simple".
Trigger Word: photorealistic (range 1-5)
2026.06.02更新:基于 "Anima base1.0"尝试性的进行全量微调,搭配我发布的lora模型,效果更佳。
数据量大约4000~5000张图。
所有的测试工作均使用“anima-turbo-lora-v0.2”、“8steps”、“er_sde”、“simple”完成。
提示词:photorealistic(可以设置的范围 1~5 )
————————————————————————————————————————————————————————————————————————————————
The following information is outdated.
以下为过时的信息。
This is a realistic photography cosplay portrait model.
"γ-00006" is a test model that used approximately 1 million photographs for training.For usage instructions, please refer to the previous version description.
I rented an RTX 4090 for approximately 1000 hours.. Due to limited funds, I may not be able to update it further.(In reality, there were approximately 1,000 hours of completely failed training, bringing the total time to over 2,000 hours.)
Since training is mostly done at low resolution, there may be some issues with human anatomy.
"γ-00006"是测试型号,使用了大约1百万张素材培训。使用方法参照之前的版本说明。租了大概1000小时4090,资金有限,后续可能不会更新。(实际上还有大约1000小时完全失败的训练,总时间超过了2000小时。)
大部分时间在低分辨率下训练,所以部分结构可能会有一些问题。
based on "Noob V1.1".
"β0.1" is an unfinished version r