一个可以让你人物不要停下来的lora,1.1版!🏃♀️🏃♀️🏃♀️
增加了速度线的效果,同时较高权重可能会更加拉大头身比,比较需要使用到高清修复,你可以根据需求选择是1.1版还是1.0版.
权重0.6~0.8,可以根据适配的人物lora进行调低或者调高.
触发词为:
kqtlt,
如果体型不是很壮硕的话,可以尝试在正向tag添加 muscle man, 来增加体型.
似乎使用1boy可以更好的提升体型,但大模型下,1girl的脸可能会更精致,你可以看情况选择.(っ °Д °;)っ
推荐使用偏正方形的分辨率,如768x768.
Description
FAQ
Comments (7)
holy...oh...
I recommend adding a stronger DETAIL
It is also recommended to add a stronger joke
Finally, slam the (work to the author)!(叩きつけてやれ)
this should be called "typical Instagram woman"
Hello, Mind if I ask about the training parameters ? what about the dataset?
I would like to train an animation grid of this lora but I want to see how you did it first if you dont mind
你可以通过"additonal Networks"这个插件看到我的lora里的具体训练参数.
ss_unet_lr: "0.00015" ss_text_encoder_lr: "3e-05" ss_optimizer: "lion" rep:"18"
数据集我不方便给你,不过你可以在pixiv上找到这些图,这些图有专门的tag,叫"走るガンガー",你可以在里面找到这种类型的图片.
The following is from machine translation:
You can see the specific training parameters in my lora through the "additonal Networks" plugin.
ss_unet_lr: "0.00015" ss_text_encoder_lr: "3e-05" ss_optimizer: "lion" rep:"18"
I'm not at liberty to give you the dataset, but you can find these graphs on pixiv, which have a special tag called "走るガンガー", and you can find these types of images in there.
{
"ss_batch_size_per_device": "5",
"ss_bucket_info": "{\"buckets\": {\"0\": {\"resolution\": [640, 896], \"count\": 18}, \"1\": {\"resolution\": [704, 832], \"count\": 18}, \"2\": {\"resolution\": [768, 768], \"count\": 36}, \"3\": {\"resolution\": [832, 704], \"count\": 108}, \"4\": {\"resolution\": [896, 640], \"count\": 90}, \"5\": {\"resolution\": [1024, 576], \"count\": 18}}, \"mean_img_ar_error\": 0.03347638038189571}",
"ss_bucket_no_upscale": "False",
"ss_cache_latents": "True",
"ss_caption_dropout_every_n_epochs": "0",
"ss_caption_dropout_rate": "0.0",
"ss_caption_tag_dropout_rate": "0.0",
"ss_clip_skip": "2",
"ss_color_aug": "False",
"ss_enable_bucket": "True",
"ss_epoch": "8",
"ss_face_crop_aug_range": "None",
"ss_flip_aug": "False",
"ss_full_fp16": "False",
"ss_gradient_accumulation_steps": "1",
"ss_gradient_checkpointing": "False",
"ss_keep_tokens": "1",
"ss_learning_rate": "0.0001",
"ss_lowram": "False",
"ss_lr_scheduler": "cosine_with_restarts",
"ss_lr_warmup_steps": "0",
"ss_max_bucket_reso": "1024",
"ss_max_grad_norm": "1.0",
"ss_max_token_length": "225",
"ss_max_train_steps": "600",
"ss_min_bucket_reso": "256",
"ss_min_snr_gamma": "None",
"ss_mixed_precision": "fp16",
"ss_network_alpha": "64.0",
"ss_network_dim": "64",
"ss_network_module": "networks.lora",
"ss_new_sd_model_hash": "32dc308ba9c67ff8780167708727e28809282077f1205e3a8a87add6654657c0",
"ss_noise_offset": "0.1",
"ss_num_batches_per_epoch": "60",
"ss_num_epochs": "10",
"ss_num_reg_images": "0",
"ss_num_train_images": "288",
"ss_optimizer": "lion_pytorch.lion_pytorch.Lion",
"ss_output_name": "kqtltV1.1",
"ss_prior_loss_weight": "1.0",
"ss_random_crop": "False",
"ss_reg_dataset_dirs": "{}",
"ss_resolution": "(768, 768)",
"ss_sd_model_hash": "ab21ba3c",
"ss_sd_model_name": "novelailatest-pruned.ckpt",
"ss_sd_scripts_commit_hash": "d52c524fc2942c053cf37c648188502a3a26df1b",
"ss_seed": "1337",
"ss_session_id": "3462523485",
"ss_shuffle_caption": "True",
"ss_steps": "480",
"ss_text_encoder_lr": "3e-05",
"ss_total_batch_size": "5",
"ss_training_comment": "None",
"ss_training_finished_at": "1684463673.5874798",
"ss_training_started_at": "1684462946.5511777",
"ss_unet_lr": "0.00015",
"ss_v2": "False"
}
@9_days thanks
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.