"Luan Fanghua" is based on a fusion of RealVisXL_V4.0 and RongHua4.5 (1%), followed by multiple rounds of training over two months using a carefully selected dataset.
All tags in the "Luan Fanghua" dataset come from GPT-4-o and the F16/florence2-large-ft-gufeng_v3 · Hugging Face model fine-tuned by me on Hugging Face.
If you are interested in fine-tuning Florence2, you can refer to the Florence2 fine-tuning script I wrote: facok/florence2-ft-simple: finetune your florence2 model easy (github.com).
乱芳华基于 RealVisXL_V4.0 和 RongHua4.5 (1%)的融合，然后在两个月时间内以精选训练集进行了多轮的训练。
乱芳华训练集的所有tag来自GPT4-o和由我微调的 F16/florence2-large-ft-gufeng_v3 · Hugging Face 模型
如果你对微调Florence2有兴趣，可以参考我写的Florence2微调脚本facok/florence2-ft-simple: finetune your florence2 model easy (github.com)
Description
FAQ
Comments (7)
sweet breezy cool. thank you~
it is different but small faces need work, they look like they came from robot chicken, sort of cartoon like. Steampunk looks good. But up close its rather blurry. There is a new model called redsea2 - super sharp, I wish this model had that clarity.
This model is excellent for generating references for cinematic lighting and composition. It responds well to prompts about basic lighting forms and ratios, providing a nice color palette for further visual development.
能出个Hyper急速版本吗
看了一下例图，发现主要特点是不能涩涩，残念
该模型对国风、古风十分友好，特别是在古建筑的理解效果蓬勃大气，时尚风格展示也相当给力，谢谢作者的杰出贡献
6,希望出个NSFW版本
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.