Hugging Face's logo

FFusion
/
FFUSION.ai-Text-Encoder-LyCORIS-SD-2.1

English
di.ffusion.ai
stable-diffusion
LyCORIS
LoRA
Model card Files Community
FFUSION.ai-Text-Encoder-LyCORIS-SD-2.1 / README.md
idlebg's picture
idlebg
Update README.md
e1d8f96
preview code
|
raw
history blame
16.6 kB
metadata
license: creativeml-openrail-m
language:
  - en
tags:
  - di.ffusion.ai
  - stable-diffusion
  - LyCORIS
  - LoRA

textenc2.jpg

Model Card for di.FFUSION.ai Text Encoder - SD 2.1 LyCORIS

di.FFUSION.ai-tXe-FXAA Trained on "121361" images.

Enhance your model's quality and sharpness using your own pre-trained Unet.

Screenshot_1282.jpg

The text encoder (without UNET) is wrapped in LyCORIS. Optimizer: torch.optim.adamw.AdamW(weight_decay=0.01, betas=(0.9, 0.99))

Network dimension/rank: 768.0 Alpha: 768.0 Module: lycoris.kohya {'conv_dim': '256', 'conv_alpha': '256', 'algo': 'loha'}

Large size due to Lyco CONV 256

textenco2.jpg

This is a heavy experimental version we used to test even with sloppy captions (quick WD tags and terrible clip), yet the results were satisfying.

Note: This is not the text encoder used in the official FFUSION AI model.

SAMPLES

Available also at https://civitai.com/models/83622

image.png xyz_grid-0069-3538254854.png xyz_grid-0090-2371661606.png xyz_grid-0133-887882152.png

For a1111 Install https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris

Download di.FFUSION.ai-tXe-FXAA to /models/Lycoris

Option1:

Insert lyco:di.FFUSION.ai-tXe-FXAA:1.0 to prompt No need to split Unet and Text Enc as its only TX encoder there.

You can go up to 2x weights

Option2: If you need it always ON (ex run a batch from txt file) then you can go to settings / Quicksettings list

image.png

add sd_lyco

restart and you should have a drop-down now 🤟 🥃

image.png

Table of Contents

Model Details

Model Description

di.FFUSION.ai-tXe-FXAA Trained on "121361" images.

Enhance your model's quality and sharpness using your own pre-trained Unet.

The text encoder (without UNET) is wrapped in LyCORIS. Optimizer: torch.optim.adamw.AdamW(weight_decay=0.01, betas=(0.9, 0.99))

Network dimension/rank: 768.0 Alpha: 768.0 Module: lycoris.kohya {'conv_dim': '256', 'conv_alpha': '256', 'algo': 'loha'}

Large size due to Lyco CONV 256

This is a heavy experimental version we used to test even with sloppy captions (quick WD tags and terrible clip), yet the results were satisfying.

Note: This is not the text encoder used in the official FFUSION AI model.

  • Developed by: FFusion.ai
  • Shared by [Optional]: idle stoev
  • Model type: Language model
  • Language(s) (NLP): en
  • License: creativeml-openrail-m
  • Parent Model: More information needed
  • Resources for more information: More information needed

Uses

Direct Use

The text encoder (without UNET) is wrapped in LyCORIS. Optimizer: torch.optim.adamw.AdamW(weight_decay=0.01, betas=(0.9, 0.99))

Network dimension/rank: 768.0 Alpha: 768.0 Module: lycoris.kohya {'conv_dim': '256', 'conv_alpha': '256', 'algo': 'loha'}

Large size due to Lyco CONV 256

Bias, Risks, and Limitations

Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups.

Recommendations

Training Details

Training Data

Trained on "121361" images.

ss_caption_tag_dropout_rate: "0.0", ss_multires_noise_discount: "0.3", ss_mixed_precision: "bf16", ss_text_encoder_lr: "1e-07", ss_keep_tokens: "3", ss_network_args: "{"conv_dim": "256", "conv_alpha": "256", "algo": "loha"}", ss_caption_dropout_rate: "0.02", ss_flip_aug: "False", ss_learning_rate: "2e-07", ss_sd_model_name: "stabilityai/stable-diffusion-2-1-base", ss_max_grad_norm: "1.0", ss_num_epochs: "2", ss_gradient_checkpointing: "False", ss_face_crop_aug_range: "None", ss_epoch: "2", ss_num_train_images: "121361", ss_color_aug: "False", ss_gradient_accumulation_steps: "1", ss_total_batch_size: "100", ss_prior_loss_weight: "1.0", ss_training_comment: "None", ss_network_dim: "768", ss_output_name: "FusionaMEGA1tX", ss_max_bucket_reso: "1024", ss_network_alpha: "768.0", ss_steps: "2444", ss_shuffle_caption: "True", ss_training_finished_at: "1684158038.0763328", ss_min_bucket_reso: "256", ss_noise_offset: "0.09", ss_enable_bucket: "True", ss_batch_size_per_device: "20", ss_max_train_steps: "2444", ss_network_module: "lycoris.kohya",

Training Procedure

Preprocessing

"{"buckets": {"0": {"resolution": [192, 256], "count": 1}, "1": {"resolution": [192, 320], "count": 1}, "2": {"resolution": [256, 384], "count": 1}, "3": {"resolution": [256, 512], "count": 1}, "4": {"resolution": [384, 576], "count": 2}, "5": {"resolution": [384, 640], "count": 2}, "6": {"resolution": [384, 704], "count": 1}, "7": {"resolution": [384, 1088], "count": 15}, "8": {"resolution": [448, 448], "count": 5}, "9": {"resolution": [448, 576], "count": 1}, "10": {"resolution": [448, 640], "count": 1}, "11": {"resolution": [448, 768], "count": 1}, "12": {"resolution": [448, 832], "count": 1}, "13": {"resolution": [448, 1088], "count": 25}, "14": {"resolution": [448, 1216], "count": 1}, "15": {"resolution": [512, 640], "count": 2}, "16": {"resolution": [512, 768], "count": 10}, "17": {"resolution": [512, 832], "count": 3}, "18": {"resolution": [512, 896], "count": 1525}, "19": {"resolution": [512, 960], "count": 2}, "20": {"resolution": [512, 1024], "count": 665}, "21": {"resolution": [512, 1088], "count": 8}, "22": {"resolution": [576, 576], "count": 5}, "23": {"resolution": [576, 768], "count": 1}, "24": {"resolution": [576, 832], "count": 667}, "25": {"resolution": [576, 896], "count": 9601}, "26": {"resolution": [576, 960], "count": 872}, "27": {"resolution": [576, 1024], "count": 17}, "28": {"resolution": [640, 640], "count": 3}, "29": {"resolution": [640, 768], "count": 7}, "30": {"resolution": [640, 832], "count": 608}, "31": {"resolution": [640, 896], "count": 90}, "32": {"resolution": [704, 640], "count": 1}, "33": {"resolution": [704, 704], "count": 11}, "34": {"resolution": [704, 768], "count": 1}, "35": {"resolution": [704, 832], "count": 1}, "36": {"resolution": [768, 640], "count": 225}, "37": {"resolution": [768, 704], "count": 6}, "38": {"resolution": [768, 768], "count": 74442}, "39": {"resolution": [832, 576], "count": 23784}, "40": {"resolution": [832, 640], "count": 554}, "41": {"resolution": [896, 512], "count": 1235}, "42": {"resolution": [896, 576], "count": 50}, "43": {"resolution": [896, 640], "count": 88}, "44": {"resolution": [960, 512], "count": 165}, "45": {"resolution": [960, 576], "count": 5246}, "46": {"resolution": [1024, 448], "count": 5}, "47": {"resolution": [1024, 512], "count": 1187}, "48": {"resolution": [1024, 576], "count": 40}, "49": {"resolution": [1088, 384], "count": 70}, "50": {"resolution": [1088, 448], "count": 36}, "51": {"resolution": [1088, 512], "count": 3}, "52": {"resolution": [1216, 448], "count": 36}, "53": {"resolution": [1344, 320], "count": 29}, "54": {"resolution": [1536, 384], "count": 1}}, "mean_img_ar_error": 0.01693107810697896}",

Speeds, Sizes, Times

ss_resolution: "(768, 768)", ss_v2: "True", ss_cache_latents: "False", ss_unet_lr: "2e-07", ss_num_reg_images: "0", ss_max_token_length: "225", ss_lr_scheduler: "linear", ss_reg_dataset_dirs: "{}", ss_lr_warmup_steps: "303", ss_num_batches_per_epoch: "1222", ss_lowram: "False", ss_multires_noise_iterations: "None", ss_optimizer: "torch.optim.adamw.AdamW(weight_decay=0.01,betas=(0.9, 0.99))",

Evaluation

Testing Data, Factors & Metrics

Testing Data

More information needed

Factors

More information needed

Metrics

More information needed

Results

More information needed

Model Examination

More information needed

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

  • Hardware Type: 8xA100
  • Hours used: 64
  • Cloud Provider: CoreWeave
  • Compute Region: US Main
  • Carbon Emitted: 6.72

Technical Specifications [optional]

Model Architecture and Objective

Enhance your model's quality and sharpness using your own pre-trained Unet.

Compute Infrastructure

More information needed

Hardware

8xA100

Software

Fully trained only with Kohya S & Shih-Ying Yeh (Kohaku-BlueLeaf) https://arxiv.org/abs/2108.06098

Citation

BibTeX:

More information needed

APA:

@misc{LyCORIS, author = "Shih-Ying Yeh (Kohaku-BlueLeaf), Yu-Guan Hsieh, Zhidong Gao", title = "LyCORIS - Lora beYond Conventional methods, Other Rank adaptation Implementations for Stable diffusion", howpublished = "\url{https://github.com/KohakuBlueleaf/LyCORIS}", month = "March", year = "2023" }

Glossary [optional]

More information needed

More Information [optional]

More information needed

Model Card Authors [optional]

idle stoev

Model Card Contact

di@ffusion.ai

How to Get Started with the Model

Use the code below to get started with the model.

Click to expand

For a1111 Install https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris

Download di.FFUSION.ai-tXe-FXAA to /models/Lycoris

Option1:

Insert lyco:di.FFUSION.ai-tXe-FXAA:1.0 to prompt No need to split Unet and Text Enc as its only TX encoder there.

You can go up to 2x weights

Option2: If you need it always ON (ex run a batch from txt file) then you can go to settings / Quicksettings list

image.png

add sd_lyco

restart and you should have a drop-down now 🤟 🥃

image.png