README.md · AiAF/Lightsource-0Lightsource-OLS_PonyXL.safetensors at 84a0dc51278b095bbf22c85c913d0d602079d7a2

metadata

license: cc0-1.0
datasets:
  - AiAF/Lightsource-0Lightsource-OLS_PonyXL-LoRA_Dataset
pipeline_tag: text-to-image
tags:
  - art

Model Description

Style LoRA based on Lightsource's Style. ( https://x.com/0Lightsource)

Please let me know your thoughts!

License: [Cretive Commons Zero UniversAL]
Finetuned from model [optional]: [ponyDiffusionV6XL_v6StartWithThisOne.safetensors (https://civitai.com/models/257749?modelVersionId=290640) (https://huggingface.co/AiAF/ponyDiffusionV6XL_v6StartWithThisOne.safetensors)]

Model Sources [optional]

Civitai Source: [https://civitai.com/models/553464?modelVersionId=615931]

Uses

Stable Diffusion Model indtended to work with PonyDiffsuionV6 and models finetuned off of it (models like AutismMix for example)

Trigger Words:

lightsource (Artist), @0Lightsource, OLS (Artist)

Out-of-Scope Use

Will not work well, if at all, with non-PonyV6 models or non-PonyV6-finetunes. Will not work at all witl SDXL 1.0, SD 1.5, etc.

[More Information Needed]

Bias, Risks, and Limitations

[Simi-important note: During the image tagging process, I accidentally mispelled Marasaki's name as "Marusaki" and didn't catch the mistake until towards the end of the training process. Unfortunately this means his appearance will only be accurately generated if its spelled "Maru" instead of the correct "Mara". I may correct this via training a new version in the future.]

Recommendations

None

How to Get Started with the Model

If on a local A1111 set up, use the standard Lora:[name-of-LoRA-Goes-here]:[1] prompt. If you're using this on an external service like Civitai then it should be self explanitory.

[More Information Needed]

Training Details

Training Data

[https://huggingface.co/datasets/AiAF/Lightsource-0Lightsource-OLS_PonyXL-LoRA_Dataset]

Interrogation Data

{ "ss_epoch": "10", "ss_bucket_no_upscale": "False", "ss_total_batch_size": "25", "ss_num_batches_per_epoch": "69", "ss_new_vae_hash": "63aeecb90ff7bc1c115395962d3e803571385b61938377bc7089b36e81e92e2e", "ss_tag_frequency": "{"meta_lat.json": {}}", "ss_min_snr_gamma": "3", "ss_caption_dropout_every_n_epochs": "0", "ss_sd_model_hash": "e577480d", "ss_max_token_length": "225", "ss_shuffle_caption": "False", "ss_seed": "4479", "ss_sd_model_name": "v6.safetensors", "ss_reg_dataset_dirs": "{}", "ss_flip_aug": "False", "ss_lr_warmup_steps": "0", "ss_resolution": "(1024, 1024)", "ss_caption_dropout_rate": "0.0", "ss_gradient_checkpointing": "True", "ss_bucket_info": "{"buckets": {"0": {"resolution": [320, 1024], "count": 4}, "1": {"resolution": [384, 1024], "count": 2}, "2": {"resolution": [448, 1024], "count": 16}, "3": {"resolution": [512, 1024], "count": 32}, "4": {"resolution": [576, 1024], "count": 72}, "5": {"resolution": [640, 1024], "count": 126}, "6": {"resolution": [704, 1024], "count": 240}, "7": {"resolution": [768, 1024], "count": 182}, "8": {"resolution": [832, 1024], "count": 164}, "9": {"resolution": [896, 1024], "count": 76}, "10": {"resolution": [960, 1024], "count": 38}, "11": {"resolution": [1024, 320], "count": 2}, "12": {"resolution": [1024, 384], "count": 4}, "13": {"resolution": [1024, 448], "count": 2}, "14": {"resolution": [1024, 512], "count": 2}, "15": {"resolution": [1024, 576], "count": 26}, "16": {"resolution": [1024, 640], "count": 42}, "17": {"resolution": [1024, 704], "count": 84}, "18": {"resolution": [1024, 768], "count": 72}, "19": {"resolution": [1024, 832], "count": 52}, "20": {"resolution": [1024, 896], "count": 42}, "21": {"resolution": [1024, 960], "count": 34}, "22": {"resolution": [1024, 1024], "count": 32}}, "mean_img_ar_error": 0.0}", "ss_full_fp16": "False", "ss_scale_weight_norms": "None", "ss_mixed_precision": "fp16", "ss_max_grad_norm": "0", "ss_enable_bucket": "True", "ss_network_dropout": "None", "ss_training_comment": "None", "ss_training_finished_at": "1720056618.6869428", "ss_multires_noise_iterations": "6", "ss_random_crop": "False", "ss_num_epochs": "10", "ss_num_reg_images": "0", "ss_network_dim": "32", "ss_network_args": "{"conv_dim": "8", "conv_alpha": "1"}", "ss_num_train_images": "1346", "ss_gradient_accumulation_steps": "1", "ss_face_crop_aug_range": "None", "ss_lowram": "False", "ss_vae_name": "sdxl_vae.safetensors", "ss_clip_skip": "None", "ss_max_bucket_reso": "None", "sshs_model_hash": "86886e99d8a83793fe63cc21287344330858888423015fd998cc133c69a18862", "ss_batch_size_per_device": "25", "ss_v2": "False", "ss_unet_lr": "None", "ss_keep_tokens": "0", "ss_color_aug": "False", "ss_noise_offset": "None", "ss_optimizer": "transformers.optimization.Adafactor(scale_parameter=False,relative_step=False,warmup_init=False)", "ss_caption_tag_dropout_rate": "0.0", "ss_base_model_version": "sdxl_base_v0-9", "ss_zero_terminal_snr": "False", "ss_max_train_steps": "690", "ss_multires_noise_discount": "0.3", "ss_learning_rate": "0.001", "ss_adaptive_noise_scale": "None", "ss_network_module": "networks.lora", "ss_steps": "690", "ss_vae_hash": "d636e597", "ss_training_started_at": "1720053253.286365", "ss_sd_scripts_commit_hash": "05811296f6dc987f67f194689e106a326017b9d4", "ss_min_bucket_reso": "None", "ss_output_name": "Lightsource | @0Lightsource | OLS[PonyXL]", "ss_network_alpha": "32", "ss_prior_loss_weight": "1.0", "ss_lr_scheduler": "constant", "ss_new_sd_model_hash": "67ab2fd8ec439a89b3fedb15cc65f54336af163c7eb5e4f2acc98f090a29b0b3", "ss_text_encoder_lr": "None", "ss_cache_latents": "False", "sshs_legacy_hash": "76a48373", "ss_dataset_dirs": "{"meta_lat.json": {"n_repeats": 2, "img_count": 673}}", "ss_session_id": "2268693806" }

[SDXL / PonyXL]

Hardware

[Nvidia T4, L4, or A100]

Software

[https://github.com/Linaqruf/kohya-trainer/blob/main/kohya-LoRA-dreambooth.ipynb]

Note: Repo is in Archived Read-only state as of May 14th, 2024. Can still be git cloned but may be depricated in some areas now that it's not being actively maintained. Contact me if you have any questions. I may be able to assist if you run into any problmems (please provide stack traces)