Upload 7 files

README.md

@@ -1,12 +1,12 @@
----
-title: Convert Repo To Safetensors Sdxl Lora
-emoji: 🏢
-colorFrom: indigo
-colorTo: red
-sdk: gradio
-sdk_version: 4.39.0
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: Convert diffusers SDXL LoRA repo to WebUI single Safetensors
+emoji: 🐶
+colorFrom: yellow
+colorTo: red
+sdk: gradio
+sdk_version: 4.38.1
+app_file: app.py
+pinned: false
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,21 @@
+import gradio as gr
+import os
+from convert_repo_to_safetensors_sdxl_lora_gr import convert_repo_to_safetensors_sdxl_lora_multi
+
+css = """"""
+
+with gr.Blocks(theme="NoCrypt/miku@>=1.2.2", css=css) as demo:
+    with gr.Column(): 
+        repo_id = gr.Textbox(label="Repo ID", placeholder="author/model", value="", lines=1)
+        run_button = gr.Button(value="Convert")
+        st_file = gr.Files(label="Output", interactive=False)
+
+    gr.on(
+        triggers=[repo_id.submit, run_button.click],
+        fn=convert_repo_to_safetensors_sdxl_lora_multi,
+        inputs=[repo_id, st_file],
+        outputs=[st_file],
+    )
+
+demo.queue()
+demo.launch()

convert_repo_to_safetensors_sdxl_lora.py

@@ -0,0 +1,74 @@
+# Script for converting a Hugging Face Diffusers trained SDXL LoRAs to Kohya format
+# This means that you can input your diffusers-trained LoRAs and
+# Get the output to work with WebUIs such as AUTOMATIC1111, ComfyUI, SD.Next and others.
+
+# To get started you can find some cool `diffusers` trained LoRAs such as this cute Corgy
+# https://huggingface.co/ignasbud/corgy_dog_LoRA/, download its `pytorch_lora_weights.safetensors` file
+# and run the script:
+# python convert_diffusers_sdxl_lora_to_webui.py --input_lora pytorch_lora_weights.safetensors --output_lora corgy.safetensors
+# now you can use corgy.safetensors in your WebUI of choice!
+
+# To train your own, here are some diffusers training scripts and utils that you can use and then convert:
+# LoRA Ease - no code SDXL Dreambooth LoRA trainer: https://huggingface.co/spaces/multimodalart/lora-ease
+# Dreambooth Advanced Training Script - state of the art techniques such as pivotal tuning and prodigy optimizer:
+# - Script: https://github.com/huggingface/diffusers/blob/main/examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py
+# - Colab (only on Pro): https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_Dreambooth_LoRA_advanced_example.ipynb
+# Canonical diffusers training scripts:
+# - Script: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_sdxl.py
+# - Colab (runs on free tier): https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb
+
+import argparse
+import os
+
+from safetensors.torch import load_file, save_file
+from diffusers.utils import convert_all_state_dict_to_peft, convert_state_dict_to_kohya
+from pathlib import Path
+
+def convert_and_save(input_lora, output_lora=None):
+    if output_lora is None:
+        base_name = os.path.splitext(input_lora)[0]
+        output_lora = f"{base_name}_webui.safetensors"
+
+    diffusers_state_dict = load_file(input_lora)
+    peft_state_dict = convert_all_state_dict_to_peft(diffusers_state_dict)
+    kohya_state_dict = convert_state_dict_to_kohya(peft_state_dict)
+    save_file(kohya_state_dict, output_lora)
+
+
+def download_repo_lora(repo_id, local_file):
+    from huggingface_hub import hf_hub_download, HfApi
+    lora_filename = "pytorch_lora_weights.safetensors"
+    lora_path = Path(lora_filename)
+    api = HfApi()
+    try:
+        if not api.file_exists(repo_id=repo_id, filename=lora_filename):
+            print(f"Error: This repo isn't diffusers LoRA repo: {repo_id}. ")
+            return None
+        if lora_path.exists():
+            print(f"Error: Download file already exists: {lora_filename}. ")
+            return None
+        hf_hub_download(repo_id=repo_id, filename=lora_filename, local_dir=".")
+        if lora_path.exists(): lora_path.rename(Path(local_file))
+    except Exception as e:
+        print(f"Error: Failed to download from {repo_id}. ")
+        return local_file
+
+
+def convert_repo_to_safetensors_sdxl_lora(repo_id):
+    download_filename = f"{repo_id.split('/')[0]}_{repo_id.split('/')[-1]}_diffusers.safetensors"
+    output_filename = f"{repo_id.split('/')[0]}_{repo_id.split('/')[-1]}_webui.safetensors"
+    download_repo_lora(repo_id, download_filename)
+    convert_and_save(download_filename, output_filename)
+    return output_filename
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Convert LoRA model to PEFT and then to Kohya format from Repo.")
+    parser.add_argument("--repo_id", type=str, required=True, help="URL to the Repo of input LoRA model in the diffusers format.")
+
+    args = parser.parse_args()
+
+    convert_repo_to_safetensors_sdxl_lora(args.repo_id)
+
+
+# Usage: python convert_repo_to_safetensors_sdxl_lora.py --repo_id nroggendorff/zelda-lora

convert_repo_to_safetensors_sdxl_lora_gr.py

@@ -0,0 +1,82 @@
+# Script for converting a Hugging Face Diffusers trained SDXL LoRAs to Kohya format
+# This means that you can input your diffusers-trained LoRAs and
+# Get the output to work with WebUIs such as AUTOMATIC1111, ComfyUI, SD.Next and others.
+
+# To get started you can find some cool `diffusers` trained LoRAs such as this cute Corgy
+# https://huggingface.co/ignasbud/corgy_dog_LoRA/, download its `pytorch_lora_weights.safetensors` file
+# and run the script:
+# python convert_diffusers_sdxl_lora_to_webui.py --input_lora pytorch_lora_weights.safetensors --output_lora corgy.safetensors
+# now you can use corgy.safetensors in your WebUI of choice!
+
+# To train your own, here are some diffusers training scripts and utils that you can use and then convert:
+# LoRA Ease - no code SDXL Dreambooth LoRA trainer: https://huggingface.co/spaces/multimodalart/lora-ease
+# Dreambooth Advanced Training Script - state of the art techniques such as pivotal tuning and prodigy optimizer:
+# - Script: https://github.com/huggingface/diffusers/blob/main/examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py
+# - Colab (only on Pro): https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_Dreambooth_LoRA_advanced_example.ipynb
+# Canonical diffusers training scripts:
+# - Script: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_sdxl.py
+# - Colab (runs on free tier): https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb
+
+import argparse
+import os
+
+from safetensors.torch import load_file, save_file
+from diffusers.utils import convert_all_state_dict_to_peft, convert_state_dict_to_kohya
+from pathlib import Path
+import gradio as gr
+
+def convert_and_save(input_lora, output_lora=None):
+    if output_lora is None:
+        base_name = os.path.splitext(input_lora)[0]
+        output_lora = f"{base_name}_webui.safetensors"
+
+    diffusers_state_dict = load_file(input_lora)
+    peft_state_dict = convert_all_state_dict_to_peft(diffusers_state_dict)
+    kohya_state_dict = convert_state_dict_to_kohya(peft_state_dict)
+    save_file(kohya_state_dict, output_lora)
+
+
+def download_repo_lora(repo_id, local_file, progress=gr.Progress(track_tqdm=True)):
+    from huggingface_hub import hf_hub_download, HfApi
+    lora_filename = "pytorch_lora_weights.safetensors"
+    lora_path = Path(lora_filename)
+    api = HfApi()
+    try:
+        if not api.file_exists(repo_id=repo_id, filename=lora_filename):
+            print(f"Error: This repo isn't diffusers LoRA repo: {repo_id}. ")
+            return None
+        if lora_path.exists():
+            print(f"Error: Download file already exists: {lora_filename}. ")
+            return None
+        hf_hub_download(repo_id=repo_id, filename=lora_filename, local_dir=".")
+        if lora_path.exists(): lora_path.rename(Path(local_file))
+    except Exception as e:
+        print(f"Error: Failed to download from {repo_id}. ")
+        return local_file
+
+
+def convert_repo_to_safetensors_sdxl_lora(repo_id, progress=gr.Progress(track_tqdm=True)):
+    download_filename = f"{repo_id.split('/')[0]}_{repo_id.split('/')[-1]}_diffusers.safetensors"
+    output_filename = f"{repo_id.split('/')[0]}_{repo_id.split('/')[-1]}_webui.safetensors"
+    download_repo_lora(repo_id, download_filename)
+    convert_and_save(download_filename, output_filename)
+    return output_filename
+
+
+def convert_repo_to_safetensors_sdxl_lora_multi(repo_id, files, progress=gr.Progress(track_tqdm=True)):
+    file = convert_repo_to_safetensors_sdxl_lora(repo_id)
+    if not files: files = []
+    files.append(file)
+    return gr.update(value=files)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Convert LoRA model to PEFT and then to Kohya format from Repo.")
+    parser.add_argument("--repo_id", type=str, required=True, help="URL to the Repo of input LoRA model in the diffusers format.")
+
+    args = parser.parse_args()
+
+    convert_repo_to_safetensors_sdxl_lora(args.repo_id)
+
+
+# Usage: python convert_repo_to_safetensors_sdxl_lora.py --repo_id nroggendorff/zelda-lora

local/convert_repo_to_safetensors_sdxl_lora.py

@@ -0,0 +1,74 @@
+# Script for converting a Hugging Face Diffusers trained SDXL LoRAs to Kohya format
+# This means that you can input your diffusers-trained LoRAs and
+# Get the output to work with WebUIs such as AUTOMATIC1111, ComfyUI, SD.Next and others.
+
+# To get started you can find some cool `diffusers` trained LoRAs such as this cute Corgy
+# https://huggingface.co/ignasbud/corgy_dog_LoRA/, download its `pytorch_lora_weights.safetensors` file
+# and run the script:
+# python convert_diffusers_sdxl_lora_to_webui.py --input_lora pytorch_lora_weights.safetensors --output_lora corgy.safetensors
+# now you can use corgy.safetensors in your WebUI of choice!
+
+# To train your own, here are some diffusers training scripts and utils that you can use and then convert:
+# LoRA Ease - no code SDXL Dreambooth LoRA trainer: https://huggingface.co/spaces/multimodalart/lora-ease
+# Dreambooth Advanced Training Script - state of the art techniques such as pivotal tuning and prodigy optimizer:
+# - Script: https://github.com/huggingface/diffusers/blob/main/examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py
+# - Colab (only on Pro): https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_Dreambooth_LoRA_advanced_example.ipynb
+# Canonical diffusers training scripts:
+# - Script: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_sdxl.py
+# - Colab (runs on free tier): https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb
+
+import argparse
+import os
+
+from safetensors.torch import load_file, save_file
+from diffusers.utils import convert_all_state_dict_to_peft, convert_state_dict_to_kohya
+from pathlib import Path
+
+def convert_and_save(input_lora, output_lora=None):
+    if output_lora is None:
+        base_name = os.path.splitext(input_lora)[0]
+        output_lora = f"{base_name}_webui.safetensors"
+
+    diffusers_state_dict = load_file(input_lora)
+    peft_state_dict = convert_all_state_dict_to_peft(diffusers_state_dict)
+    kohya_state_dict = convert_state_dict_to_kohya(peft_state_dict)
+    save_file(kohya_state_dict, output_lora)
+
+
+def download_repo_lora(repo_id, local_file):
+    from huggingface_hub import hf_hub_download, HfApi
+    lora_filename = "pytorch_lora_weights.safetensors"
+    lora_path = Path(lora_filename)
+    api = HfApi()
+    try:
+        if not api.file_exists(repo_id=repo_id, filename=lora_filename):
+            print(f"Error: This repo isn't diffusers LoRA repo: {repo_id}. ")
+            return None
+        if lora_path.exists():
+            print(f"Error: Download file already exists: {lora_filename}. ")
+            return None
+        hf_hub_download(repo_id=repo_id, filename=lora_filename, local_dir=".")
+        if lora_path.exists(): lora_path.rename(Path(local_file))
+    except Exception as e:
+        print(f"Error: Failed to download from {repo_id}. ")
+        return local_file
+
+
+def convert_repo_to_safetensors_sdxl_lora(repo_id):
+    download_filename = f"{repo_id.split('/')[0]}_{repo_id.split('/')[-1]}_diffusers.safetensors"
+    output_filename = f"{repo_id.split('/')[0]}_{repo_id.split('/')[-1]}_webui.safetensors"
+    download_repo_lora(repo_id, download_filename)
+    convert_and_save(download_filename, output_filename)
+    return output_filename
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Convert LoRA model to PEFT and then to Kohya format from Repo.")
+    parser.add_argument("--repo_id", type=str, required=True, help="URL to the Repo of input LoRA model in the diffusers format.")
+
+    args = parser.parse_args()
+
+    convert_repo_to_safetensors_sdxl_lora(args.repo_id)
+
+
+# Usage: python convert_repo_to_safetensors_sdxl_lora.py --repo_id nroggendorff/zelda-lora

local/requirements.txt

@@ -0,0 +1,5 @@
+accelerate
+diffusers
+torch
+transformers
+peft

requirements.txt

@@ -0,0 +1,5 @@
+accelerate
+diffusers
+torch
+transformers
+peft