Please add AI tag if you post images on media!
Since my English is not very good, I used machine translation to introduce all the models. There may be some discrepancies, please forgive me.
The training set tag of the model was all marked manually, and the background was removed. The collection of pictures and the editing training of marks were all completed by me independently.
A slightly lower resolution is recommended for drawing and reuse (Hires. fix)
Make a fix for better quality
Recommended parameters are as follows
parameters:
score_9,score_8_up,score_7_up,score_6_up,score_5_up,score_4_up,
han fu,
lora weight recommendation :0.8
Negative prompt: boring_e621_fluffyrock_v4_sdxl,
About 古装/汉服/Han Fu_SDXL_pony bate1.0:
21 images trained on Pony Diffusion V6 XL model.LoRA
About 古装/汉服/Han Fu_SDXL_pony bate 1.5:
26 images trained on Pony Diffusion V6 XL model.LoRA
Beta version 1.5, if lora is used alone to produce straight pictures, the effect is good, and the artist label can greatly erase the excess wrinkles and clothing errors. If the character lora is used, the model requirements of the character lora may be higher, but the artist label can greatly alleviate the model requirements.
About 古装/汉服/Han Fu_SDXL_pony bate 2.0 Lork:
26 images trained on Pony Diffusion V6 XL model.LyCORIS-Lokr
This is a version of FP8. Please enable support for FP8 lora
The grid dimension is trained using lokr, why this version is 2.0 compared to lora has unparalleled memory utilization advantages as long as 6.9mb can achieve the effect of previous hundreds of mb size models, and the effect is much better than lora with fewer errors and better logic.
However, due to the low resolution of the training set, there are some strange shadows and lines, which will be gradually fixed in the beta
This version is a preview and is expected to continue to be updated, but I need time to update the training set, originally in order to use Dan on the furry test on the human body has a better effect, recommend using the indigo XL model for drawing
If you find any problems, please comment.
Welcome to post any images.
Description
This is a version of FP8. Please enable support for FP8 lora
The grid dimension is trained using lokr, why this version is 2.0 compared to lora has unparalleled memory utilization advantages as long as 6.9mb can achieve the effect of previous hundreds of mb size models, and the effect is much better than lora with fewer errors and better logic.