v1

README.md

@@ -1,10 +1,10 @@
 ---
 title: Huanyan Studio
-emoji: 📊
+emoji: ✨
 colorFrom: blue
 colorTo: indigo
 sdk: gradio
-sdk_version: 5.17.0
+sdk_version: 5.16.0
 app_file: app.py
 pinned: false
 license: mit

app.py

@@ -0,0 +1,7 @@
+from utils import install_packages
+
+if __name__ == "__main__":
+    install_packages()
+
+    from demo_app import demo
+    demo.queue(max_size=20).launch()

demo_app.py

@@ -0,0 +1,110 @@
+import gradio as gr
+from PIL import Image
+import torch
+from diffusers import HunyuanVideoPipeline
+# ... other imports ...
+
+# Add LORA configuration
+LORA_LIST = [
+    "Top_Off.safetensors",
+    "huanyan_helper.safetensors",
+    "huanyan_helper_alpha.safetensors",
+    "hunyuan-t-solo-v1.0.safetensors",
+    "stripe_v2.safetensors"
+]
+
+def create_advanced_settings():
+    with gr.Accordion("Advanced Settings", open=False):
+        # LORA Selection
+        lora_choices = gr.CheckboxGroup(
+            choices=LORA_LIST,
+            label="Select LORAs",
+            value=[LORA_LIST[0]]
+        )
+
+        lora_weights = {}
+        for lora in LORA_LIST:
+            lora_weights[lora] = gr.Slider(0.0, 1.0, value=0.8,
+                                        label=f"{lora} Weight")
+
+        # Resolution Settings
+        resolution = gr.Dropdown(
+            choices=["512x512", "768x768", "1024x1024"],
+            value="512x512",
+            label="Output Resolution"
+        )
+
+    return lora_choices, lora_weights, resolution
+
+def validate_image_resolution(image, resolution):
+    if image is None:
+        return
+    img = Image.open(image)
+    w, h = img.size
+    if f"{w}x{h}" != resolution:
+        raise gr.Error(f"Image resolution ({w}x{h}) must match output resolution ({resolution})")
+
+def generate_video(prompt, negative_prompt, lora_choices, lora_weights,
+                  resolution, image_input=None, steps=30):
+    # Validate image resolution if provided
+    if image_input:
+        validate_image_resolution(image_input, resolution)
+
+    # Load base model
+    pipe = HunyuanVideoPipeline.from_pretrained(
+        "Tencent-Hunyuan/Hunyuan-Video-Lite",
+        torch_dtype=torch.float16
+    ).to("cuda")
+
+    # Apply selected LORAs
+    for lora in lora_choices:
+        pipe.load_lora_weights(
+            f"TTV4ME/{lora}",
+            adapter_name="hunyuanvideo-lora",
+            weight_name=lora_weights[lora]
+        )
+
+    # Generate from image or text
+    if image_input:
+        image = Image.open(image_input).convert("RGB")
+        output = pipe.image_to_video(
+            image,
+            prompt=prompt,
+            negative_prompt=negative_prompt,
+            num_frames=24,
+            height=int(resolution.split("x")[1]),
+            width=int(resolution.split("x")[0]),
+            num_inference_steps=steps
+        )
+    else:
+        output = pipe.text_to_video(
+            prompt=prompt,
+            negative_prompt=negative_prompt,
+            height=int(resolution.split("x")[1]),
+            width=int(resolution.split("x")[0]),
+            num_inference_steps=steps
+        )
+
+    return output.video
+
+# Update interface
+with gr.Blocks() as demo:
+    with gr.Row():
+        with gr.Column():
+            prompt = gr.Textbox(label="Prompt")
+            negative_prompt = gr.Textbox(label="Negative Prompt")
+            image_input = gr.Image(label="Input Image", type="filepath")
+
+            lora_choices, lora_weights, resolution = create_advanced_settings()
+
+            generate_btn = gr.Button("Generate Video")
+
+        with gr.Column():
+            output_video = gr.Video(label="Generated Video")
+
+    generate_btn.click(
+        fn=generate_video,
+        inputs=[prompt, negative_prompt, lora_choices,
+               lora_weights, resolution, image_input],
+        outputs=output_video
+    )

packages.txt

@@ -0,0 +1,4 @@
+ffmpeg
+python3-imageio
+cmake
+libstdc++6

requirements.txt

@@ -0,0 +1,48 @@
+--extra-index-url https://download.pytorch.org/whl/cu124
+bitsandbytes
+decord
+einops
+facexlib
+ftfy
+gguf
+git+https://github.com/huggingface/accelerate.git@main#egg=accelerate
+git+https://github.com/huggingface/diffusers.git@main#egg=diffusers
+git+https://github.com/huggingface/transformers.git@main#egg=transformers
+gradio
+hf_transfer
+huggingface_hub
+imageio
+imageio-ffmpeg
+insightface
+invisible_watermark
+matplotlib
+moviepy==1.0.3
+numpy<2.0
+onnxruntime
+onnxruntime-gpu
+omegaconf
+opencv-python
+opencv-python-headless
+git+https://github.com/huggingface/optimum-quanto
+packaging
+patch_conv
+Pillow==10.2.0
+psutil
+safetensors
+scipy
+scikit-learn
+scikit-image
+scikit-video
+sentencepiece
+setuptools
+spaces
+timm
+tokenizers>=0.13.3
+torch<2.6.0,>=2.4.0
+torchao
+torchaudio
+torchsde
+torchvision
+tqdm
+wheel
+git+https://github.com/huggingface/peft.git

utils.py

@@ -0,0 +1,40 @@
+def install_packages():
+    import subprocess
+    import sys
+    import importlib
+
+    def _is_package_available(name) -> bool:
+        try:
+            importlib.import_module(name)
+            return True
+        except (ImportError, ModuleNotFoundError):
+            return False
+
+    # upgrade pip
+    subprocess.run(
+        f"{sys.executable} -m pip install --upgrade pip", shell=True, check=True
+    )
+    subprocess.run(
+        f"{sys.executable} -m pip install --upgrade ninja wheel setuptools packaging", shell=True, check=True
+    )
+
+    # install ninja
+    if not _is_package_available("ninja"):
+        subprocess.run(f"{sys.executable} -m pip install ninja nvidia-cudnn-cu12==9.1.0.70 nvidia-cublas-cu12==12.4.5.8 torch==2.5.1 --extra-index-url https://download.pytorch.org/whl/cu124", shell=True, check=True)
+
+    # install flash attention
+    if not _is_package_available("flash_attn"):
+        subprocess.run(
+            f"{sys.executable} -m pip install -v -U flash-attention --no-build-isolation",
+            env={"MAX_JOBS": "1"},
+            shell=True,
+            check=True
+        )
+
+    # install xformers
+    if not _is_package_available("xformers"):
+        subprocess.run(
+            f"{sys.executable} -m pip install -v -U xformers nvidia-cudnn-cu12==9.1.0.70 nvidia-cublas-cu12==12.4.5.8 torch==2.5.1 --extra-index-url https://download.pytorch.org/whl/cu124",
+            shell=True,
+            check=True
+        )