Initial app

.flake8

@@ -0,0 +1,21 @@
+[flake8]
+max-line-length = 100
+ignore =
+    # whitespace before ':' (conflicted with Black)
+    E203,
+    # ambiguous variable name
+    E741,
+    # ‘from module import *’ used; unable to detect undefined names
+    F403,
+    # name may be undefined, or defined from star imports: module
+    F405,
+    # redefinition of unused name from line N
+    F811,
+    # undefined name
+    F821,
+    # line break before binary operator
+    W503,
+    # line break after binary operator
+    W504
+# module imported but unused
+per-file-ignores = __init__.py: F401

.gitignore

@@ -0,0 +1,55 @@
+# Compiled Object files
+*.slo
+*.lo
+*.o
+*.cuo
+
+# Compiled Dynamic libraries
+*.so
+*.dll
+*.dylib
+
+# Compiled Static libraries
+*.lai
+*.la
+*.a
+*.lib
+
+# Compiled python
+*.pyc
+__pycache__
+
+# Compiled MATLAB
+*.mex*
+
+# IPython notebook checkpoints
+.ipynb_checkpoints
+
+# Editor temporaries
+*.swp
+*~
+
+# Sublime Text settings
+*.sublime-workspace
+*.sublime-project
+
+# Eclipse Project settings
+*.*project
+.settings
+
+# QtCreator files
+*.user
+
+# VSCode files
+.vscode
+
+# IDEA files
+.idea
+
+# OSX dir files
+.DS_Store
+
+# Android files
+.gradle
+*.iml
+local.properties

README.md

@@ -1,5 +1,5 @@
 ---
-title: Nova D48w1024 Sdxl1024
+title: nova-d48w1024-sdxl1024
 emoji: 📉
 colorFrom: purple
 colorTo: gray
@@ -8,7 +8,7 @@ sdk_version: 5.9.1
 app_file: app.py
 pinned: false
 license: apache-2.0
-short_description: 'NOVA-0.6B Text-to-Image 1024x1024 '
+short_description: 'NOVA Text-to-Image APP'
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,7 +1,155 @@
+# Copyright (c) 2024-present, BAAI. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+##############################################################################
+"""NOVA T2I application."""
+
+import argparse
+import os
+
 import gradio as gr
+import numpy as np
+import spaces
+import torch
+
+from diffnext.pipelines import NOVAPipeline
+from diffnext.utils import export_to_image
+
+# Switch to the allocator optimized for dynamic shape.
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
+
+
+def parse_args():
+    """Parse arguments."""
+    parser = argparse.ArgumentParser(description="Serve NOVA T2I application")
+    parser.add_argument("--model", default="", help="model path")
+    parser.add_argument("--device", type=int, default=0, help="device index")
+    parser.add_argument("--precision", default="float16", help="compute precision")
+    return parser.parse_args()
+
+
+@spaces.GPU()
+def generate_image4(
+    prompt,
+    negative_prompt,
+    seed,
+    randomize_seed,
+    guidance_scale,
+    num_inference_steps,
+    num_diffusion_steps,
+    progress=gr.Progress(track_tqdm=True),
+):
+    """Generate 4 images."""
+    args = locals()
+    seed = np.random.randint(2147483647) if randomize_seed else seed
+    device = getattr(pipe, "_offload_device", pipe.device)
+    generator = torch.Generator(device=device).manual_seed(seed)
+    images = pipe(generator=generator, num_images_per_prompt=4, **args).images
+    return [export_to_image(image, quality=95) for image in images] + [seed]
+
+
+css = """#col-container {margin: 0 auto; max-width: 1366px}"""
+title = "Autoregressive Video Generation without Vector Quantization"
+abbr = "<strong>NO</strong>n-quantized <strong>V</strong>ideo <strong>A</strong>utoregressive"
+header = (
+    "<div align='center'>"
+    "<h2>Autoregressive Video Generation without Vector Quantization</h2>"
+    "<h3><a href='https://arxiv.org/abs/2412.14169' target='_blank' rel='noopener'>[paper]</a>"
+    "<a href='https://github.com/baaivision/NOVA' target='_blank' rel='noopener'>[code]</a></h3>"
+    "</div>"
+)
+header2 = f"<div align='center'><h3>🖼️ A {abbr} model for continuous visual generation</h3></div>"
+
+examples = [
+    "a selfie of an old man with a white beard.",
+    "a woman with long hair next to a luminescent bird.",
+    "a digital artwork of a cat styled in a whimsical fashion. The overall vibe is quirky and artistic.",  # noqa
+    "a shiba inu wearing a beret and black turtleneck.",
+    "a beautiful afghan women by red hair and green eyes.",
+    "beautiful fireworks in the sky with red, white and blue.",
+    "A dragon perched majestically on a craggy, smoke-wreathed mountain.",
+    "A photo of llama wearing sunglasses standing on the deck of a spaceship with the Earth in the background.",  # noqa
+    "Two pandas in fluffy slippers and bathrobes, lazily munching on bamboo.",
+]
+
+
+if __name__ == "__main__":
+    args = parse_args()
+
+    device_type = "cuda" if torch.cuda.is_available() else "cpu"
+    device, dtype = torch.device(device_type, args.device), getattr(torch, args.precision.lower())
+    pipe = NOVAPipeline.from_pretrained(args.model, torch_dtype=dtype).to(device)
+
+    # Main Application.
+    app = gr.Blocks(css=css, theme="origin").__enter__()
+    container = gr.Column(elem_id="col-container").__enter__()
+    _, main_row = gr.Markdown(header), gr.Row().__enter__()
+
+    # Input.
+    input_col = gr.Column().__enter__()
+    prompt = gr.Text(
+        label="Prompt",
+        placeholder="Describe the video you want to generate",
+        value="a shiba inu wearing a beret and black turtleneck.",
+        lines=5,
+    )
+    negative_prompt = gr.Text(
+        label="Negative Prompt",
+        placeholder="Describe what you don't want in the image",
+        value="low quality, deformed, distorted, disfigured, fused fingers, bad anatomy, weird hand",  # noqa
+        lines=5,
+    )
+    # fmt: off
+    adv_opt = gr.Accordion("Advanced Options", open=True).__enter__()
+    seed = gr.Slider(label="Seed", maximum=2147483647, step=1, value=0)
+    randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+    guidance_scale = gr.Slider(label="Guidance scale", minimum=1, maximum=10, step=0.1, value=5)
+    with gr.Row():
+        num_inference_steps = gr.Slider(label="Inference steps", minimum=1, maximum=128, step=1, value=64) # noqa
+        num_diffusion_steps = gr.Slider(label="Diffusion steps", minimum=1, maximum=50, step=1, value=25)  # noqa
+    adv_opt.__exit__()
+    generate = gr.Button("Generate Image", variant="primary", size="lg")
+    input_col.__exit__()
+    # fmt: on
+
+    # Results.
+    result_col, _ = gr.Column().__enter__(), gr.Markdown(header2)
+    with gr.Row():
+        result1 = gr.Image(label="Result1", show_label=False)
+        result2 = gr.Image(label="Result2", show_label=False)
+    with gr.Row():
+        result3 = gr.Image(label="Result3", show_label=False)
+        result4 = gr.Image(label="Result4", show_label=False)
+    result_col.__exit__(), main_row.__exit__()
 
-def greet(name):
-    return "Hello " + name + "!!"
+    # Examples.
+    with gr.Row():
+        gr.Examples(examples=examples, inputs=[prompt])
 
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch()
+    # Events.
+    container.__exit__()
+    gr.on(
+        triggers=[generate.click, prompt.submit, negative_prompt.submit],
+        fn=generate_image4,
+        inputs=[
+            prompt,
+            negative_prompt,
+            seed,
+            randomize_seed,
+            guidance_scale,
+            num_inference_steps,
+            num_diffusion_steps,
+        ],
+        outputs=[result1, result2, result3, result4, seed],
+    )
+    app.__exit__(), app.launch()

requirements.txt

@@ -0,0 +1,6 @@
+torch
+diffusers
+transformers
+accelerate
+imageio[ffmpeg]
+git+https://github.com/baaivision/NOVA.git