【Hackathon 5th No.84】新增模型视频生成模型MS-Image2Video+MS-Vid2Vid-XL

ppdiffusers/examples/inference/I2VGen-XL-gradio/I2VGen-XL.gradio.py

@@ -0,0 +1,75 @@
+import gradio as gr
+import paddle
+from ppdiffusers import ImgToVideoSDPipeline, VideoToVideoModelscopePipeline
+from ppdiffusers.utils import export_to_video, load_image
+
+image_to_video_pipe = ImgToVideoSDPipeline.from_pretrained(
+    "Yang-Changhui/img-to-video-paddle", paddle_dtype=paddle.float32
+)
+
+video_to_video_pipe = VideoToVideoModelscopePipeline.from_pretrained("Yang-Changhui/video-to-video-paddle")
+
+
+def upload_file(file):
+    return file.name
+
+
+def image_to_video(image_in):
+    if image_in is None:
+        raise gr.Error('请上传图片或等待图片上传完成')
+    image_in = load_image(image_in)
+    output_video_frames = image_to_video_pipe(image_in).frames
+    output_video_path = export_to_video(output_video_frames, "img2video_test.mp4")
+    print(output_video_path)
+    return output_video_path
+
+
+def video_to_video(video_in, text_in):
+    output_video_frames = video_to_video_pipe(prompt=text_in, video_path=video_in).frames
+    output_video_path = export_to_video(output_video_frames, "video2video_test.mp4")
+    print(output_video_path)
+    return output_video_path
+
+
+with gr.Blocks() as demo:
+    gr.Markdown(
+        """<center><font size=7>I2VGen-XL</center>
+        <left><font size=3>I2VGen-XL可以根据用户输入的静态图像和文本生成目标接近、语义相同的视频，生成的视频具高清(1280 * 720)、宽屏(16:9)、时序连贯、质感好等特点。</left>
+
+        <left><font size=3>I2VGen-XL can generate videos with similar contents and semantics based on user input static images and text. The generated videos have characteristics such as high-definition (1280 * 720), widescreen (16:9), coherent timing, and good texture.</left>
+        """
+    )
+    with gr.Blocks():
+        gr.Markdown(
+            """<left><font size=3>步骤1：选择合适的图片进行上传 (建议图片比例为1：1)，然后点击“生成视频”，得到满意的视频后进行下一步。”</left>
+
+        <left><font size=3>Step 1:Select the image to upload (it is recommended that the image ratio is 1:1), and then click on “Generate Video” to obtain a generated video before proceeding to the next step.</left>"""
+        )
+        with gr.Row():
+            with gr.Column():
+                image_in = gr.Image(label="图片输入", type="filepath", interactive=False, elem_id="image-in", height=300)
+                with gr.Row():
+                    upload_image = gr.UploadButton("上传图片", file_types=["image"], file_count="single")
+                    image_submit = gr.Button("生成视频🎬")
+            with gr.Column():
+                video_out_1 = gr.Video(label='生成的视频', elem_id='video-out_1', interactive=False, height=300)
+        gr.Markdown(
+            """<left><font size=3>步骤2：补充对视频内容的英文文本描述，然后点击“生成高分辨率视频”，视频生成大致需要2分钟。”</left>
+
+        <left><font size=3>Step 1:Add the English text description of the video you want to generate, and then click on “Generate high-resolution video”. The video generation will take about 2 minutes..</left>"""
+        )
+        with gr.Row():
+            with gr.Column():
+                text_in = gr.Textbox(label="文本描述", lines=2, elem_id="text-in")
+                video_submit = gr.Button("生成高分辨率视频🎬")
+            with gr.Column():
+                paddle.device.cuda.empty_cache()
+                video_out_2 = gr.Video(label='生成的视频', elem_id='video-out_2', interactive=False, height=300)
+    gr.Markdown("<left><font size=2>注：如果生成的视频无法播放，请尝试升级浏览器或使用chrome浏览器。</left>")
+
+    upload_image.upload(upload_file, upload_image, image_in, queue=False)
+    image_submit.click(fn=image_to_video, inputs=[image_in], outputs=[video_out_1])
+
+    video_submit.click(fn=video_to_video, inputs=[video_out_1, text_in], outputs=[video_out_2])
+
+demo.queue(status_update_rate=1, api_open=False).launch(share=False, show_error=True)

ppdiffusers/ppdiffusers/__init__.py

@@ -58,20 +58,25 @@
     from .models import (
         AsymmetricAutoencoderKL,
         AutoencoderKL,
+        AutoencoderKL_imgtovideo,
         ControlNetModel,
+        GaussianDiffusion,
+        GaussianDiffusion_SDEdit,
         LitEma,
         LVDMAutoencoderKL,
         LVDMUNet3DModel,
         ModelMixin,
         MultiAdapter,
         PriorTransformer,
+        STUNetModel,
         T2IAdapter,
         T5FilmDecoder,
         Transformer2DModel,
         UNet1DModel,
         UNet2DConditionModel,
         UNet2DModel,
         UNet3DConditionModel,
+        Vid2VidSTUNet,
         VQModel,
     )
     from .optimization import (
@@ -170,6 +175,7 @@
         IFInpaintingSuperResolutionPipeline,
         IFPipeline,
         IFSuperResolutionPipeline,
+        ImgToVideoSDPipeline,
         KandinskyCombinedPipeline,
         KandinskyImg2ImgCombinedPipeline,
         KandinskyImg2ImgPipeline,
@@ -234,6 +240,7 @@
         VersatileDiffusionImageVariationPipeline,
         VersatileDiffusionPipeline,
         VersatileDiffusionTextToImagePipeline,
+        VideoToVideoModelscopePipeline,
         VideoToVideoSDPipeline,
         VQDiffusionPipeline,
     )

ppdiffusers/ppdiffusers/models/__init__.py

@@ -23,14 +23,21 @@
 if is_paddle_available():
     from .adapter import MultiAdapter, T2IAdapter
     from .autoencoder_asym_kl import AsymmetricAutoencoderKL
+
     from .autoencoder_kl import AutoencoderKL
     from .controlnet import ControlNetModel
     from .dual_transformer_2d import DualTransformer2DModel
     from .ema import LitEma
     from .lvdm_unet_3d import LVDMUNet3DModel
     from .lvdm_vae import LVDMAutoencoderKL
     from .modeling_utils import ModelMixin
+    from .modelscope_autoencoder_img2vid import AutoencoderKL_imgtovideo
+    from .modelscope_gaussian_diffusion import GaussianDiffusion
+    from .modelscope_gaussion_sdedit import GaussianDiffusion_SDEdit
+    from .modelscope_st_unet import STUNetModel
+    from .modelscope_st_unet_video2video import Vid2VidSTUNet
     from .prior_transformer import PriorTransformer
+
     from .t5_film_transformer import T5FilmDecoder
     from .transformer_2d import Transformer2DModel
     from .unet_1d import UNet1DModel