Verified: a year ago
SafeTensor
The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.
IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
v2
使用5000+图片训练了33000+步的flux尝试版本，使用了lokr，数据集不露点，尽可能地选择了较为丰满诱人的图片，但由于训练次数还有欠缺，所以还有待继续训练。
训练的时候，数据打标使用了关键词打标和自然语言打标两种合并起来，所以理论上使用哪种都会有不错的效果，但还是推荐使用comfyui的clip文本编码FLUX节点，同时输入两种。具体可以参考我上传的图片中的工作流
An attempted version of FLUX with 33,000+ steps was trained using 5,000+ images, using lokr, the dataset is not revealing, and more voluptuous and tantalizing images were chosen as much as possible, but the number of trainings is still lacking, so there is still some training to be done.
When training, data marking using keyword marking and natural language marking two combined, so theoretically using either will have good results, but still recommended to use comfyui's clip text encoding FLUX node, at the same time input two. Specifically, you can refer to the workflow in the image I uploaded
v1
基于4000+写真照片练的丹，权重太高似乎还是会崩坏，建议权重0.7-0.8
推荐采样方法使用Euler a或者DPM++ SDE Karras
Dan based on 4000+ portrait photos, it seems that the weight is too high, it will still collapse, the recommended weight is 0.7-0.8
The recommended sampling method is Euler a or DPM++ SDE Karras