Spaces:
Paused
Paused
Create app-backup1.py
Browse files- app-backup1.py +679 -0
app-backup1.py
ADDED
|
@@ -0,0 +1,679 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from gradio_toggle import Toggle
|
| 3 |
+
import torch
|
| 4 |
+
from huggingface_hub import snapshot_download
|
| 5 |
+
from transformers import pipeline
|
| 6 |
+
|
| 7 |
+
from xora.models.autoencoders.causal_video_autoencoder import CausalVideoAutoencoder
|
| 8 |
+
from xora.models.transformers.transformer3d import Transformer3DModel
|
| 9 |
+
from xora.models.transformers.symmetric_patchifier import SymmetricPatchifier
|
| 10 |
+
from xora.schedulers.rf import RectifiedFlowScheduler
|
| 11 |
+
from xora.pipelines.pipeline_xora_video import XoraVideoPipeline
|
| 12 |
+
from transformers import T5EncoderModel, T5Tokenizer
|
| 13 |
+
from xora.utils.conditioning_method import ConditioningMethod
|
| 14 |
+
from pathlib import Path
|
| 15 |
+
import safetensors.torch
|
| 16 |
+
import json
|
| 17 |
+
import numpy as np
|
| 18 |
+
import cv2
|
| 19 |
+
from PIL import Image
|
| 20 |
+
import tempfile
|
| 21 |
+
import os
|
| 22 |
+
import gc
|
| 23 |
+
from openai import OpenAI
|
| 24 |
+
import re
|
| 25 |
+
|
| 26 |
+
# Load system prompts
|
| 27 |
+
system_prompt_t2v = """๋น์ ์ ๋น๋์ค ์์ฑ์ ์ํ ํ๋กฌํํธ ์ ๋ฌธ๊ฐ์
๋๋ค.
|
| 28 |
+
์ฃผ์ด์ง ํ๋กฌํํธ๋ฅผ ๋ค์ ๊ตฌ์กฐ์ ๋ง๊ฒ ๊ฐ์ ํด์ฃผ์ธ์:
|
| 29 |
+
1. ์ฃผ์ ๋์์ ๋ช
ํํ ํ ๋ฌธ์ฅ์ผ๋ก ์์
|
| 30 |
+
2. ๊ตฌ์ฒด์ ์ธ ๋์๊ณผ ์ ์ค์ฒ๋ฅผ ์๊ฐ ์์๋๋ก ์ค๋ช
|
| 31 |
+
3. ์บ๋ฆญํฐ/๊ฐ์ฒด์ ์ธ๋ชจ๋ฅผ ์์ธํ ๋ฌ์ฌ
|
| 32 |
+
4. ๋ฐฐ๊ฒฝ๊ณผ ํ๊ฒฝ ์ธ๋ถ ์ฌํญ์ ๊ตฌ์ฒด์ ์ผ๋ก ํฌํจ
|
| 33 |
+
5. ์นด๋ฉ๋ผ ๊ฐ๋์ ์์ง์์ ๋ช
์
|
| 34 |
+
6. ์กฐ๋ช
๊ณผ ์์์ ์์ธํ ์ค๋ช
|
| 35 |
+
7. ๋ณํ๋ ๊ฐ์์ค๋ฌ์ด ์ฌ๊ฑด์ ์์ฐ์ค๋ฝ๊ฒ ํฌํจ
|
| 36 |
+
๋ชจ๋ ์ค๋ช
์ ํ๋์ ์์ฐ์ค๋ฌ์ด ๋ฌธ๋จ์ผ๋ก ์์ฑํ๊ณ ,
|
| 37 |
+
์ดฌ์ ๊ฐ๋
์ด ์ดฌ์ ๋ชฉ๋ก์ ์ค๋ช
ํ๋ ๊ฒ์ฒ๋ผ ๊ตฌ์ฒด์ ์ด๊ณ ์๊ฐ์ ์ผ๋ก ์์ฑํ์ธ์.
|
| 38 |
+
200๋จ์ด๋ฅผ ๋์ง ์๋๋ก ํ๋, ์ต๋ํ ์์ธํ๊ฒ ์์ฑํ์ธ์."""
|
| 39 |
+
|
| 40 |
+
system_prompt_i2v = """๋น์ ์ ์ด๋ฏธ์ง ๊ธฐ๋ฐ ๋น๋์ค ์์ฑ์ ์ํ ํ๋กฌํํธ ์ ๋ฌธ๊ฐ์
๋๋ค.
|
| 41 |
+
์ฃผ์ด์ง ํ๋กฌํํธ๋ฅผ ๋ค์ ๊ตฌ์กฐ์ ๋ง๊ฒ ๊ฐ์ ํด์ฃผ์ธ์:
|
| 42 |
+
1. ์ฃผ์ ๋์์ ๋ช
ํํ ํ ๋ฌธ์ฅ์ผ๋ก ์์
|
| 43 |
+
2. ๊ตฌ์ฒด์ ์ธ ๋์๊ณผ ์ ์ค์ฒ๋ฅผ ์๊ฐ ์์๋๋ก ์ค๋ช
|
| 44 |
+
3. ์บ๋ฆญํฐ/๊ฐ์ฒด์ ์ธ๋ชจ๋ฅผ ์์ธํ ๋ฌ์ฌ
|
| 45 |
+
4. ๋ฐฐ๊ฒฝ๊ณผ ํ๊ฒฝ ์ธ๋ถ ์ฌํญ์ ๊ตฌ์ฒด์ ์ผ๋ก ํฌํจ
|
| 46 |
+
5. ์นด๋ฉ๋ผ ๊ฐ๋์ ์์ง์์ ๋ช
์
|
| 47 |
+
6. ์กฐ๋ช
๊ณผ ์์์ ์์ธํ ์ค๋ช
|
| 48 |
+
7. ๋ณํ๋ ๊ฐ์์ค๋ฌ์ด ์ฌ๊ฑด์ ์์ฐ์ค๋ฝ๊ฒ ํฌํจ
|
| 49 |
+
๋ชจ๋ ์ค๋ช
์ ํ๋์ ์์ฐ์ค๋ฌ์ด ๋ฌธ๋จ์ผ๋ก ์์ฑํ๊ณ ,
|
| 50 |
+
์ดฌ์ ๊ฐ๋
์ด ์ดฌ์ ๋ชฉ๋ก์ ์ค๋ช
ํ๋ ๊ฒ์ฒ๋ผ ๊ตฌ์ฒด์ ์ด๊ณ ์๊ฐ์ ์ผ๋ก ์์ฑํ์ธ์.
|
| 51 |
+
200๋จ์ด๋ฅผ ๋์ง ์๋๋ก ํ๋, ์ต๋ํ ์์ธํ๊ฒ ์์ฑํ์ธ์."""
|
| 52 |
+
|
| 53 |
+
# Load Hugging Face token if needed
|
| 54 |
+
hf_token = os.getenv("HF_TOKEN")
|
| 55 |
+
openai_api_key = os.getenv("OPENAI_API_KEY")
|
| 56 |
+
client = OpenAI(api_key=openai_api_key)
|
| 57 |
+
|
| 58 |
+
# Initialize translation pipeline
|
| 59 |
+
translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")
|
| 60 |
+
|
| 61 |
+
# Korean text detection function
|
| 62 |
+
def contains_korean(text):
|
| 63 |
+
korean_pattern = re.compile('[ใฑ-ใ
ใ
-ใ
ฃ๊ฐ-ํฃ]')
|
| 64 |
+
return bool(korean_pattern.search(text))
|
| 65 |
+
|
| 66 |
+
def translate_korean_prompt(prompt):
|
| 67 |
+
"""
|
| 68 |
+
Translate Korean prompt to English if Korean text is detected
|
| 69 |
+
"""
|
| 70 |
+
if contains_korean(prompt):
|
| 71 |
+
translated = translator(prompt)[0]['translation_text']
|
| 72 |
+
print(f"Original Korean prompt: {prompt}")
|
| 73 |
+
print(f"Translated English prompt: {translated}")
|
| 74 |
+
return translated
|
| 75 |
+
return prompt
|
| 76 |
+
|
| 77 |
+
def enhance_prompt(prompt, type="t2v"):
|
| 78 |
+
system_prompt = system_prompt_t2v if type == "t2v" else system_prompt_i2v
|
| 79 |
+
messages = [
|
| 80 |
+
{"role": "system", "content": system_prompt},
|
| 81 |
+
{"role": "user", "content": prompt},
|
| 82 |
+
]
|
| 83 |
+
|
| 84 |
+
try:
|
| 85 |
+
response = client.chat.completions.create(
|
| 86 |
+
model="gpt-4-1106-preview",
|
| 87 |
+
messages=messages,
|
| 88 |
+
max_tokens=2000,
|
| 89 |
+
)
|
| 90 |
+
enhanced_prompt = response.choices[0].message.content.strip()
|
| 91 |
+
|
| 92 |
+
print("\n=== ํ๋กฌํํธ ์ฆ๊ฐ ๊ฒฐ๊ณผ ===")
|
| 93 |
+
print("Original Prompt:")
|
| 94 |
+
print(prompt)
|
| 95 |
+
print("\nEnhanced Prompt:")
|
| 96 |
+
print(enhanced_prompt)
|
| 97 |
+
print("========================\n")
|
| 98 |
+
|
| 99 |
+
return enhanced_prompt
|
| 100 |
+
except Exception as e:
|
| 101 |
+
print(f"Error during prompt enhancement: {e}")
|
| 102 |
+
return prompt
|
| 103 |
+
|
| 104 |
+
def update_prompt_t2v(prompt, enhance_toggle):
|
| 105 |
+
return update_prompt(prompt, enhance_toggle, "t2v")
|
| 106 |
+
|
| 107 |
+
def update_prompt_i2v(prompt, enhance_toggle):
|
| 108 |
+
return update_prompt(prompt, enhance_toggle, "i2v")
|
| 109 |
+
|
| 110 |
+
def update_prompt(prompt, enhance_toggle, type="t2v"):
|
| 111 |
+
if enhance_toggle:
|
| 112 |
+
return enhance_prompt(prompt, type)
|
| 113 |
+
return prompt
|
| 114 |
+
|
| 115 |
+
# Set model download directory within Hugging Face Spaces
|
| 116 |
+
model_path = "asset"
|
| 117 |
+
if not os.path.exists(model_path):
|
| 118 |
+
snapshot_download(
|
| 119 |
+
"Lightricks/LTX-Video", local_dir=model_path, repo_type="model", token=hf_token
|
| 120 |
+
)
|
| 121 |
+
|
| 122 |
+
# Global variables to load components
|
| 123 |
+
vae_dir = Path(model_path) / "vae"
|
| 124 |
+
unet_dir = Path(model_path) / "unet"
|
| 125 |
+
scheduler_dir = Path(model_path) / "scheduler"
|
| 126 |
+
|
| 127 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 128 |
+
|
| 129 |
+
def load_vae(vae_dir):
|
| 130 |
+
vae_ckpt_path = vae_dir / "vae_diffusion_pytorch_model.safetensors"
|
| 131 |
+
vae_config_path = vae_dir / "config.json"
|
| 132 |
+
with open(vae_config_path, "r") as f:
|
| 133 |
+
vae_config = json.load(f)
|
| 134 |
+
vae = CausalVideoAutoencoder.from_config(vae_config)
|
| 135 |
+
vae_state_dict = safetensors.torch.load_file(vae_ckpt_path)
|
| 136 |
+
vae.load_state_dict(vae_state_dict)
|
| 137 |
+
return vae.to(device=device, dtype=torch.bfloat16)
|
| 138 |
+
|
| 139 |
+
def load_unet(unet_dir):
|
| 140 |
+
unet_ckpt_path = unet_dir / "unet_diffusion_pytorch_model.safetensors"
|
| 141 |
+
unet_config_path = unet_dir / "config.json"
|
| 142 |
+
transformer_config = Transformer3DModel.load_config(unet_config_path)
|
| 143 |
+
transformer = Transformer3DModel.from_config(transformer_config)
|
| 144 |
+
unet_state_dict = safetensors.torch.load_file(unet_ckpt_path)
|
| 145 |
+
transformer.load_state_dict(unet_state_dict, strict=True)
|
| 146 |
+
return transformer.to(device=device, dtype=torch.bfloat16)
|
| 147 |
+
|
| 148 |
+
def load_scheduler(scheduler_dir):
|
| 149 |
+
scheduler_config_path = scheduler_dir / "scheduler_config.json"
|
| 150 |
+
scheduler_config = RectifiedFlowScheduler.load_config(scheduler_config_path)
|
| 151 |
+
return RectifiedFlowScheduler.from_config(scheduler_config)
|
| 152 |
+
|
| 153 |
+
# Helper function for image processing
|
| 154 |
+
def center_crop_and_resize(frame, target_height, target_width):
|
| 155 |
+
h, w, _ = frame.shape
|
| 156 |
+
aspect_ratio_target = target_width / target_height
|
| 157 |
+
aspect_ratio_frame = w / h
|
| 158 |
+
if aspect_ratio_frame > aspect_ratio_target:
|
| 159 |
+
new_width = int(h * aspect_ratio_target)
|
| 160 |
+
x_start = (w - new_width) // 2
|
| 161 |
+
frame_cropped = frame[:, x_start : x_start + new_width]
|
| 162 |
+
else:
|
| 163 |
+
new_height = int(w / aspect_ratio_target)
|
| 164 |
+
y_start = (h - new_height) // 2
|
| 165 |
+
frame_cropped = frame[y_start : y_start + new_height, :]
|
| 166 |
+
frame_resized = cv2.resize(frame_cropped, (target_width, target_height))
|
| 167 |
+
return frame_resized
|
| 168 |
+
|
| 169 |
+
def load_image_to_tensor_with_resize(image_path, target_height=512, target_width=768):
|
| 170 |
+
image = Image.open(image_path).convert("RGB")
|
| 171 |
+
image_np = np.array(image)
|
| 172 |
+
frame_resized = center_crop_and_resize(image_np, target_height, target_width)
|
| 173 |
+
frame_tensor = torch.tensor(frame_resized).permute(2, 0, 1).float()
|
| 174 |
+
frame_tensor = (frame_tensor / 127.5) - 1.0
|
| 175 |
+
return frame_tensor.unsqueeze(0).unsqueeze(2)
|
| 176 |
+
|
| 177 |
+
# Load models
|
| 178 |
+
vae = load_vae(vae_dir)
|
| 179 |
+
unet = load_unet(unet_dir)
|
| 180 |
+
scheduler = load_scheduler(scheduler_dir)
|
| 181 |
+
patchifier = SymmetricPatchifier(patch_size=1)
|
| 182 |
+
text_encoder = T5EncoderModel.from_pretrained(
|
| 183 |
+
"PixArt-alpha/PixArt-XL-2-1024-MS", subfolder="text_encoder"
|
| 184 |
+
).to(device)
|
| 185 |
+
tokenizer = T5Tokenizer.from_pretrained(
|
| 186 |
+
"PixArt-alpha/PixArt-XL-2-1024-MS", subfolder="tokenizer"
|
| 187 |
+
)
|
| 188 |
+
|
| 189 |
+
pipeline = XoraVideoPipeline(
|
| 190 |
+
transformer=unet,
|
| 191 |
+
patchifier=patchifier,
|
| 192 |
+
text_encoder=text_encoder,
|
| 193 |
+
tokenizer=tokenizer,
|
| 194 |
+
scheduler=scheduler,
|
| 195 |
+
vae=vae,
|
| 196 |
+
).to(device)
|
| 197 |
+
|
| 198 |
+
# Preset options for resolution and frame configuration
|
| 199 |
+
preset_options = [
|
| 200 |
+
{"label": "1216x704, 41 frames", "width": 1216, "height": 704, "num_frames": 41},
|
| 201 |
+
{"label": "1088x704, 49 frames", "width": 1088, "height": 704, "num_frames": 49},
|
| 202 |
+
{"label": "1056x640, 57 frames", "width": 1056, "height": 640, "num_frames": 57},
|
| 203 |
+
{"label": "992x608, 65 frames", "width": 992, "height": 608, "num_frames": 65},
|
| 204 |
+
{"label": "896x608, 73 frames", "width": 896, "height": 608, "num_frames": 73},
|
| 205 |
+
{"label": "896x544, 81 frames", "width": 896, "height": 544, "num_frames": 81},
|
| 206 |
+
{"label": "832x544, 89 frames", "width": 832, "height": 544, "num_frames": 89},
|
| 207 |
+
{"label": "800x512, 97 frames", "width": 800, "height": 512, "num_frames": 97},
|
| 208 |
+
{"label": "768x512, 97 frames", "width": 768, "height": 512, "num_frames": 97},
|
| 209 |
+
{"label": "800x480, 105 frames", "width": 800, "height": 480, "num_frames": 105},
|
| 210 |
+
{"label": "736x480, 113 frames", "width": 736, "height": 480, "num_frames": 113},
|
| 211 |
+
{"label": "704x480, 121 frames", "width": 704, "height": 480, "num_frames": 121},
|
| 212 |
+
{"label": "704x448, 129 frames", "width": 704, "height": 448, "num_frames": 129},
|
| 213 |
+
{"label": "672x448, 137 frames", "width": 672, "height": 448, "num_frames": 137},
|
| 214 |
+
{"label": "640x416, 153 frames", "width": 640, "height": 416, "num_frames": 153},
|
| 215 |
+
{"label": "672x384, 161 frames", "width": 672, "height": 384, "num_frames": 161},
|
| 216 |
+
{"label": "640x384, 169 frames", "width": 640, "height": 384, "num_frames": 169},
|
| 217 |
+
{"label": "608x384, 177 frames", "width": 608, "height": 384, "num_frames": 177},
|
| 218 |
+
{"label": "576x384, 185 frames", "width": 576, "height": 384, "num_frames": 185},
|
| 219 |
+
{"label": "608x352, 193 frames", "width": 608, "height": 352, "num_frames": 193},
|
| 220 |
+
{"label": "576x352, 201 frames", "width": 576, "height": 352, "num_frames": 201},
|
| 221 |
+
{"label": "544x352, 209 frames", "width": 544, "height": 352, "num_frames": 209},
|
| 222 |
+
{"label": "512x352, 225 frames", "width": 512, "height": 352, "num_frames": 225},
|
| 223 |
+
{"label": "512x352, 233 frames", "width": 512, "height": 352, "num_frames": 233},
|
| 224 |
+
{"label": "544x320, 241 frames", "width": 544, "height": 320, "num_frames": 241},
|
| 225 |
+
{"label": "512x320, 249 frames", "width": 512, "height": 320, "num_frames": 249},
|
| 226 |
+
{"label": "512x320, 257 frames", "width": 512, "height": 320, "num_frames": 257},
|
| 227 |
+
]
|
| 228 |
+
|
| 229 |
+
def preset_changed(preset):
|
| 230 |
+
if preset != "Custom":
|
| 231 |
+
selected = next(item for item in preset_options if item["label"] == preset)
|
| 232 |
+
return (
|
| 233 |
+
selected["height"],
|
| 234 |
+
selected["width"],
|
| 235 |
+
selected["num_frames"],
|
| 236 |
+
gr.update(visible=False),
|
| 237 |
+
gr.update(visible=False),
|
| 238 |
+
gr.update(visible=False),
|
| 239 |
+
)
|
| 240 |
+
else:
|
| 241 |
+
return (
|
| 242 |
+
None,
|
| 243 |
+
None,
|
| 244 |
+
None,
|
| 245 |
+
gr.update(visible=True),
|
| 246 |
+
gr.update(visible=True),
|
| 247 |
+
gr.update(visible=True),
|
| 248 |
+
)
|
| 249 |
+
|
| 250 |
+
def generate_video_from_text(
|
| 251 |
+
prompt="",
|
| 252 |
+
enhance_prompt_toggle=False,
|
| 253 |
+
negative_prompt="",
|
| 254 |
+
frame_rate=25,
|
| 255 |
+
seed=171198,
|
| 256 |
+
num_inference_steps=30,
|
| 257 |
+
guidance_scale=3,
|
| 258 |
+
height=512,
|
| 259 |
+
width=768,
|
| 260 |
+
num_frames=121,
|
| 261 |
+
progress=gr.Progress(),
|
| 262 |
+
):
|
| 263 |
+
if len(prompt.strip()) < 50:
|
| 264 |
+
raise gr.Error(
|
| 265 |
+
"ํ๋กฌํํธ๋ ์ต์ 50์ ์ด์์ด์ด์ผ ํฉ๋๋ค. ๋ ์์ธํ ์ค๋ช
์ ์ ๊ณตํด์ฃผ์ธ์.",
|
| 266 |
+
duration=5,
|
| 267 |
+
)
|
| 268 |
+
|
| 269 |
+
# Translate Korean prompts to English
|
| 270 |
+
prompt = translate_korean_prompt(prompt)
|
| 271 |
+
negative_prompt = translate_korean_prompt(negative_prompt)
|
| 272 |
+
|
| 273 |
+
sample = {
|
| 274 |
+
"prompt": prompt,
|
| 275 |
+
"prompt_attention_mask": None,
|
| 276 |
+
"negative_prompt": negative_prompt,
|
| 277 |
+
"negative_prompt_attention_mask": None,
|
| 278 |
+
"media_items": None,
|
| 279 |
+
}
|
| 280 |
+
|
| 281 |
+
generator = torch.Generator(device="cpu").manual_seed(seed)
|
| 282 |
+
|
| 283 |
+
def gradio_progress_callback(self, step, timestep, kwargs):
|
| 284 |
+
progress((step + 1) / num_inference_steps)
|
| 285 |
+
|
| 286 |
+
try:
|
| 287 |
+
with torch.no_grad():
|
| 288 |
+
images = pipeline(
|
| 289 |
+
num_inference_steps=num_inference_steps,
|
| 290 |
+
num_images_per_prompt=1,
|
| 291 |
+
guidance_scale=guidance_scale,
|
| 292 |
+
generator=generator,
|
| 293 |
+
output_type="pt",
|
| 294 |
+
height=height,
|
| 295 |
+
width=width,
|
| 296 |
+
num_frames=num_frames,
|
| 297 |
+
frame_rate=frame_rate,
|
| 298 |
+
**sample,
|
| 299 |
+
is_video=True,
|
| 300 |
+
vae_per_channel_normalize=True,
|
| 301 |
+
conditioning_method=ConditioningMethod.UNCONDITIONAL,
|
| 302 |
+
mixed_precision=True,
|
| 303 |
+
callback_on_step_end=gradio_progress_callback,
|
| 304 |
+
).images
|
| 305 |
+
except Exception as e:
|
| 306 |
+
raise gr.Error(
|
| 307 |
+
f"๋น๋์ค ์์ฑ ์ค ์ค๋ฅ๊ฐ ๋ฐ์ํ์ต๋๋ค. ๋ค์ ์๋ํด์ฃผ์ธ์. ์ค๋ฅ: {e}",
|
| 308 |
+
duration=5,
|
| 309 |
+
)
|
| 310 |
+
finally:
|
| 311 |
+
torch.cuda.empty_cache()
|
| 312 |
+
gc.collect()
|
| 313 |
+
|
| 314 |
+
output_path = tempfile.mktemp(suffix=".mp4")
|
| 315 |
+
print(images.shape)
|
| 316 |
+
video_np = images.squeeze(0).permute(1, 2, 3, 0).cpu().float().numpy()
|
| 317 |
+
video_np = (video_np * 255).astype(np.uint8)
|
| 318 |
+
height, width = video_np.shape[1:3]
|
| 319 |
+
out = cv2.VideoWriter(
|
| 320 |
+
output_path, cv2.VideoWriter_fourcc(*"mp4v"), frame_rate, (width, height)
|
| 321 |
+
)
|
| 322 |
+
for frame in video_np[..., ::-1]:
|
| 323 |
+
out.write(frame)
|
| 324 |
+
out.release()
|
| 325 |
+
del images
|
| 326 |
+
del video_np
|
| 327 |
+
torch.cuda.empty_cache()
|
| 328 |
+
return output_path
|
| 329 |
+
|
| 330 |
+
def generate_video_from_image(
|
| 331 |
+
image_path,
|
| 332 |
+
prompt="",
|
| 333 |
+
enhance_prompt_toggle=False,
|
| 334 |
+
negative_prompt="",
|
| 335 |
+
frame_rate=25,
|
| 336 |
+
seed=171198,
|
| 337 |
+
num_inference_steps=30,
|
| 338 |
+
guidance_scale=3,
|
| 339 |
+
height=512,
|
| 340 |
+
width=768,
|
| 341 |
+
num_frames=121,
|
| 342 |
+
progress=gr.Progress(),
|
| 343 |
+
):
|
| 344 |
+
print("Height: ", height)
|
| 345 |
+
print("Width: ", width)
|
| 346 |
+
print("Num Frames: ", num_frames)
|
| 347 |
+
|
| 348 |
+
if len(prompt.strip()) < 50:
|
| 349 |
+
raise gr.Error(
|
| 350 |
+
"ํ๋กฌํํธ๋ ์ต์ 50์ ์ด์์ด์ด์ผ ํฉ๋๋ค. ๋ ์์ธํ ์ค๋ช
์ ์ ๊ณตํด์ฃผ์ธ์.",
|
| 351 |
+
duration=5,
|
| 352 |
+
)
|
| 353 |
+
|
| 354 |
+
if not image_path:
|
| 355 |
+
raise gr.Error("์
๋ ฅ ์ด๋ฏธ์ง๋ฅผ ์ ๊ณตํด์ฃผ์ธ์.", duration=5)
|
| 356 |
+
|
| 357 |
+
# Translate Korean prompts to English
|
| 358 |
+
prompt = translate_korean_prompt(prompt)
|
| 359 |
+
negative_prompt = translate_korean_prompt(negative_prompt)
|
| 360 |
+
|
| 361 |
+
media_items = (
|
| 362 |
+
load_image_to_tensor_with_resize(image_path, height, width).to(device).detach()
|
| 363 |
+
)
|
| 364 |
+
|
| 365 |
+
sample = {
|
| 366 |
+
"prompt": prompt,
|
| 367 |
+
"prompt_attention_mask": None,
|
| 368 |
+
"negative_prompt": negative_prompt,
|
| 369 |
+
"negative_prompt_attention_mask": None,
|
| 370 |
+
"media_items": media_items,
|
| 371 |
+
}
|
| 372 |
+
|
| 373 |
+
generator = torch.Generator(device="cpu").manual_seed(seed)
|
| 374 |
+
|
| 375 |
+
def gradio_progress_callback(self, step, timestep, kwargs):
|
| 376 |
+
progress((step + 1) / num_inference_steps)
|
| 377 |
+
|
| 378 |
+
try:
|
| 379 |
+
with torch.no_grad():
|
| 380 |
+
images = pipeline(
|
| 381 |
+
num_inference_steps=num_inference_steps,
|
| 382 |
+
num_images_per_prompt=1,
|
| 383 |
+
guidance_scale=guidance_scale,
|
| 384 |
+
generator=generator,
|
| 385 |
+
output_type="pt",
|
| 386 |
+
height=height,
|
| 387 |
+
width=width,
|
| 388 |
+
num_frames=num_frames,
|
| 389 |
+
frame_rate=frame_rate,
|
| 390 |
+
**sample,
|
| 391 |
+
is_video=True,
|
| 392 |
+
vae_per_channel_normalize=True,
|
| 393 |
+
conditioning_method=ConditioningMethod.FIRST_FRAME,
|
| 394 |
+
mixed_precision=True,
|
| 395 |
+
callback_on_step_end=gradio_progress_callback,
|
| 396 |
+
).images
|
| 397 |
+
|
| 398 |
+
output_path = tempfile.mktemp(suffix=".mp4")
|
| 399 |
+
video_np = images.squeeze(0).permute(1, 2, 3, 0).cpu().float().numpy()
|
| 400 |
+
video_np = (video_np * 255).astype(np.uint8)
|
| 401 |
+
height, width = video_np.shape[1:3]
|
| 402 |
+
out = cv2.VideoWriter(
|
| 403 |
+
output_path, cv2.VideoWriter_fourcc(*"mp4v"), frame_rate, (width, height)
|
| 404 |
+
)
|
| 405 |
+
for frame in video_np[..., ::-1]:
|
| 406 |
+
out.write(frame)
|
| 407 |
+
out.release()
|
| 408 |
+
except Exception as e:
|
| 409 |
+
raise gr.Error(
|
| 410 |
+
f"๋น๋์ค ์์ฑ ์ค ์ค๋ฅ๊ฐ ๋ฐ์ํ์ต๋๋ค. ๋ค์ ์๋ํด์ฃผ์ธ์. ์ค๋ฅ: {e}",
|
| 411 |
+
duration=5,
|
| 412 |
+
)
|
| 413 |
+
|
| 414 |
+
finally:
|
| 415 |
+
torch.cuda.empty_cache()
|
| 416 |
+
gc.collect()
|
| 417 |
+
|
| 418 |
+
return output_path
|
| 419 |
+
|
| 420 |
+
def create_advanced_options():
|
| 421 |
+
with gr.Accordion("Step 4: Advanced Options (Optional)", open=False):
|
| 422 |
+
seed = gr.Slider(
|
| 423 |
+
label="4.1 Seed", minimum=0, maximum=1000000, step=1, value=171198
|
| 424 |
+
)
|
| 425 |
+
inference_steps = gr.Slider(
|
| 426 |
+
label="4.2 Inference Steps", minimum=1, maximum=50, step=1, value=30
|
| 427 |
+
)
|
| 428 |
+
guidance_scale = gr.Slider(
|
| 429 |
+
label="4.3 Guidance Scale", minimum=1.0, maximum=5.0, step=0.1, value=3.0
|
| 430 |
+
)
|
| 431 |
+
|
| 432 |
+
height_slider = gr.Slider(
|
| 433 |
+
label="4.4 Height",
|
| 434 |
+
minimum=256,
|
| 435 |
+
maximum=1024,
|
| 436 |
+
step=64,
|
| 437 |
+
value=512,
|
| 438 |
+
visible=False,
|
| 439 |
+
)
|
| 440 |
+
width_slider = gr.Slider(
|
| 441 |
+
label="4.5 Width",
|
| 442 |
+
minimum=256,
|
| 443 |
+
maximum=1024,
|
| 444 |
+
step=64,
|
| 445 |
+
value=768,
|
| 446 |
+
visible=False,
|
| 447 |
+
)
|
| 448 |
+
num_frames_slider = gr.Slider(
|
| 449 |
+
label="4.5 Number of Frames",
|
| 450 |
+
minimum=1,
|
| 451 |
+
maximum=200,
|
| 452 |
+
step=1,
|
| 453 |
+
value=121,
|
| 454 |
+
visible=False,
|
| 455 |
+
)
|
| 456 |
+
|
| 457 |
+
return [
|
| 458 |
+
seed,
|
| 459 |
+
inference_steps,
|
| 460 |
+
guidance_scale,
|
| 461 |
+
height_slider,
|
| 462 |
+
width_slider,
|
| 463 |
+
num_frames_slider,
|
| 464 |
+
]
|
| 465 |
+
|
| 466 |
+
# Gradio Interface Definition
|
| 467 |
+
with gr.Blocks(theme=gr.themes.Soft()) as iface:
|
| 468 |
+
with gr.Tabs():
|
| 469 |
+
# Text to Video Tab
|
| 470 |
+
with gr.TabItem("ํ
์คํธ๋ก ๋น๋์ค ๋ง๋ค๊ธฐ"):
|
| 471 |
+
with gr.Row():
|
| 472 |
+
with gr.Column():
|
| 473 |
+
txt2vid_prompt = gr.Textbox(
|
| 474 |
+
label="Step 1: ํ๋กฌํํธ ์
๋ ฅ",
|
| 475 |
+
placeholder="์์ฑํ๊ณ ์ถ์ ๋น๋์ค๋ฅผ ์ค๋ช
ํ์ธ์ (์ต์ 50์)...",
|
| 476 |
+
value="๊ฐ์ ๊ธด ๋จธ๋ฆฌ๋ฅผ ๊ฐ์ง ์ฌ์ฑ์ด ๊ธ๋ฐ์ ๊ธด ๋จธ๋ฆฌ๋ฅผ ๊ฐ์ง ๋ค๋ฅธ ์ฌ์ฑ์ ํฅํด ๋ฏธ์์ง์ต๋๋ค. ๊ฐ์ ๋จธ๋ฆฌ์ ์ฌ์ฑ์ ๊ฒ์์ ์์ผ์ ์
๊ณ ์์ผ๋ฉฐ ์ค๋ฅธ์ชฝ ๋บจ์ ์์ ์ ์ด ์์ต๋๋ค. ์นด๋ฉ๋ผ ๊ฐ๋๋ ๊ฐ์ ๋จธ๋ฆฌ ์ฌ์ฑ์ ์ผ๊ตด์ ํด๋ก์ฆ์
๋์ด ์์ต๋๋ค. ์กฐ๋ช
์ ์์ฐ์ค๋ฝ๊ณ ๋ฐ๋ปํ๋ฉฐ, ์์์์ ์ค๋ ๋ฏํ ๋ถ๋๋ฌ์ด ๋น์ด ์ฅ๋ฉด์ ๋น์ถฅ๋๋ค. ์ฅ๋ฉด์ ์ค์ ์์์ฒ๋ผ ๋ณด์
๋๋ค.",
|
| 477 |
+
lines=5,
|
| 478 |
+
)
|
| 479 |
+
txt2vid_enhance_toggle = Toggle(
|
| 480 |
+
label="ํ๋กฌํํธ ๊ฐ์ ",
|
| 481 |
+
value=False,
|
| 482 |
+
interactive=True,
|
| 483 |
+
)
|
| 484 |
+
|
| 485 |
+
txt2vid_negative_prompt = gr.Textbox(
|
| 486 |
+
label="Step 2: ๋ค๊ฑฐํฐ๋ธ ํ๋กฌํํธ ์
๋ ฅ",
|
| 487 |
+
placeholder="๋น๋์ค์์ ์ํ์ง ์๋ ์์๋ฅผ ์ค๋ช
ํ์ธ์...",
|
| 488 |
+
value="low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 489 |
+
lines=2,
|
| 490 |
+
)
|
| 491 |
+
|
| 492 |
+
txt2vid_preset = gr.Dropdown(
|
| 493 |
+
choices=[p["label"] for p in preset_options],
|
| 494 |
+
value="768x512, 97 frames",
|
| 495 |
+
label="Step 3.1: ํด์๋ ํ๋ฆฌ์
์ ํ",
|
| 496 |
+
)
|
| 497 |
+
|
| 498 |
+
txt2vid_frame_rate = gr.Slider(
|
| 499 |
+
label="Step 3.2: ํ๋ ์ ๋ ์ดํธ",
|
| 500 |
+
minimum=21,
|
| 501 |
+
maximum=30,
|
| 502 |
+
step=1,
|
| 503 |
+
value=25,
|
| 504 |
+
)
|
| 505 |
+
|
| 506 |
+
txt2vid_advanced = create_advanced_options()
|
| 507 |
+
txt2vid_generate = gr.Button(
|
| 508 |
+
"Step 5: ๋น๋์ค ์์ฑ",
|
| 509 |
+
variant="primary",
|
| 510 |
+
size="lg",
|
| 511 |
+
)
|
| 512 |
+
|
| 513 |
+
with gr.Column():
|
| 514 |
+
txt2vid_output = gr.Video(label="์์ฑ๋ ๋น๋์ค")
|
| 515 |
+
|
| 516 |
+
with gr.Row():
|
| 517 |
+
gr.Examples(
|
| 518 |
+
examples=[
|
| 519 |
+
[
|
| 520 |
+
"์ ํต์ ์ธ ๋ชฝ๊ณจ ๋๋ ์ค๋ฅผ ์
์ ์ ์ ์ฌ์ฑ์ด ์์ ํฐ์ ์ปคํผ์ ํตํด ํธ๊ธฐ์ฌ๊ณผ ๊ธด์ฅ์ด ์์ธ ํ์ ์ผ๋ก ๋ค์ฌ๋ค๋ณด๊ณ ์์ต๋๋ค. ์ฌ์ฑ์ ํฐ ๊ตฌ์ฌ๋ก ์ฅ์๋ ๋ ๊ฐ์ ๋์ ๋จธ๋ฆฌ๋ก ์คํ์ผ๋ง๋ ๊ธด ๊ฒ์ ๋จธ๋ฆฌ๋ฅผ ํ๊ณ ์์ผ๋ฉฐ, ๋์ ๋๋์ ๋๋ฉฐ ํฌ๊ฒ ๋ ์ ธ ์์ต๋๋ค. ๊ทธ๋
์ ๋๋ ์ค๋ ํ๋ คํ ๊ธ์ ์์๊ฐ ์๊ฒจ์ง ์ ๋ช
ํ ํ๋์์ด๋ฉฐ, ๋น์ทํ ๋์์ธ์ ๋จธ๋ฆฌ๋ ๋ฅผ ํ๊ณ ์์ต๋๋ค. ๋ฐฐ๊ฒฝ์ ์ ๋น๋ก์๊ณผ ํธ๊ธฐ์ฌ์ ์์๋ด๋ ๋จ์ํ ํฐ์ ์ปคํผ์
๋๋ค.",
|
| 521 |
+
"low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 522 |
+
"assets/t2v_2.mp4",
|
| 523 |
+
],
|
| 524 |
+
[
|
| 525 |
+
"๋
ธ๋์ ์ฌํท์ ์
์ ๊ธ๋ฐ ๋จธ๋ฆฌ์ ์ ์ ๋จ์๊ฐ ์ฒ์ ์์ ์ฃผ์๋ฅผ ๋๋ฌ๋ด
๋๋ค. ๊ทธ๋ ๋ฐ์ ํผ๋ถ๋ฅผ ๊ฐ์ก๊ณ ๋จธ๋ฆฌ๋ ๊ฐ์ด๋ฐ ๊ฐ๋ฅด๋ง๋ก ์คํ์ผ๋ง๋์ด ์์ต๋๋ค. ๊ทธ๋ ์ผ์ชฝ์ ๋ณด๊ณ ๋ ํ ์ค๋ฅธ์ชฝ์ ๋ณด๋ฉฐ, ๊ฐ ๋ฐฉํฅ์ ์ ์ ์์ํฉ๋๋ค. ์นด๋ฉ๋ผ๋ ๋ฎ์ ๊ฐ๋์์ ๋จ์๋ฅผ ์ฌ๋ ค๋ค๋ณด๋ฉฐ ๊ณ ์ ๋์ด ์์ต๋๋ค. ๋ฐฐ๊ฒฝ์ ์ฝ๊ฐ ํ๋ฆฟํ๋ฉฐ, ๋
น์ ๋๋ฌด๋ค๊ณผ ๋จ์์ ๋ค์์ ๋ฐ๊ฒ ๋น์น๋ ํ์์ด ๋ณด์
๋๋ค. ์กฐ๋ช
์ ์์ฐ์ค๋ฝ๊ณ ๋ฐ๋ปํ๋ฉฐ, ํ์ ๋น์ด ๋จ์์ ์ผ๊ตด์ ๊ฐ๋ก์ง๋ฅด๋ ๋ ์ฆ ํ๋ ์ด๋ฅผ ๋ง๋ญ๋๋ค. ์ฅ๋ฉด์ ์ค์ ์์์ฒ๋ผ ์ดฌ์๋์์ต๋๋ค.",
|
| 526 |
+
"low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 527 |
+
"assets/t2v_1.mp4",
|
| 528 |
+
],
|
| 529 |
+
[
|
| 530 |
+
"ํ ์ฌ์ดํด๋ฆฌ์คํธ๊ฐ ๊ตฝ์ด์ง ์ฐ๊ธธ์ ๋ฐ๋ผ ๋ฌ๋ฆฝ๋๋ค. ๊ณต๊ธฐ์ญํ์ ์ธ ์ฅ๋น๋ฅผ ์
์ ๊ทธ๋ ๊ฐํ๊ฒ ํ๋ฌ์ ๋ฐ๊ณ ์์ผ๋ฉฐ, ์ด๋ง์๋ ๋๋ฐฉ์ธ์ด ๋ฐ์ง์
๋๋ค. ์นด๋ฉ๋ผ๋ ๊ทธ์ ๊ฒฐ์ฐํ ํ์ ๊ณผ ์จ ๋งํ๋ ํ๊ฒฝ์ ๋ฒ๊ฐ์๊ฐ๋ฉฐ ๋ณด์ฌ์ค๋๋ค. ์๋๋ฌด๋ค์ด ์ค์ณ ์ง๋๊ฐ๊ณ , ํ๋์ ์ ๋ช
ํ ํ๋์์
๋๋ค. ์ด ์ฅ๋ฉด์ ํ๊ธฐ์ฐจ๊ณ ๊ฒฝ์์ ์ธ ๋ถ์๊ธฐ๋ฅผ ์์๋
๋๋ค.",
|
| 531 |
+
"low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 532 |
+
"assets/t2v_0.mp4",
|
| 533 |
+
],
|
| 534 |
+
],
|
| 535 |
+
inputs=[txt2vid_prompt, txt2vid_negative_prompt, txt2vid_output],
|
| 536 |
+
label="ํ
์คํธ-๋น๋์ค ์์ฑ ์์",
|
| 537 |
+
)
|
| 538 |
+
|
| 539 |
+
# Image to Video Tab
|
| 540 |
+
with gr.TabItem("์ด๋ฏธ์ง๋ก ๋น๋์ค ๋ง๋ค๊ธฐ"):
|
| 541 |
+
with gr.Row():
|
| 542 |
+
with gr.Column():
|
| 543 |
+
img2vid_image = gr.Image(
|
| 544 |
+
type="filepath",
|
| 545 |
+
label="Step 1: ์
๋ ฅ ์ด๋ฏธ์ง ์
๋ก๋",
|
| 546 |
+
elem_id="image_upload",
|
| 547 |
+
)
|
| 548 |
+
img2vid_prompt = gr.Textbox(
|
| 549 |
+
label="Step 2: ํ๋กฌํํธ ์
๋ ฅ",
|
| 550 |
+
placeholder="์ด๋ฏธ์ง๋ฅผ ์ด๋ป๊ฒ ์ ๋๋ฉ์ด์
ํํ ์ง ์ค๋ช
ํ์ธ์ (์ต์ 50์)...",
|
| 551 |
+
value="๊ฐ์ ๊ธด ๋จธ๋ฆฌ๋ฅผ ๊ฐ์ง ์ฌ์ฑ์ด ๊ธ๋ฐ์ ๊ธด ๋จธ๋ฆฌ๋ฅผ ๊ฐ์ง ๋ค๋ฅธ ์ฌ์ฑ์ ํฅํด ๋ฏธ์์ง์ต๋๋ค. ๊ฐ์ ๋จธ๋ฆฌ์ ์ฌ์ฑ์ ๊ฒ์์ ์์ผ์ ์
๊ณ ์์ผ๋ฉฐ ์ค๋ฅธ์ชฝ ๋บจ์ ์์ ์ ์ด ์์ต๋๋ค. ์นด๋ฉ๋ผ ๊ฐ๋๋ ๊ฐ์ ๋จธ๋ฆฌ ์ฌ์ฑ์ ์ผ๊ตด์ ํด๋ก์ฆ์
๋์ด ์์ต๋๋ค. ์กฐ๋ช
์ ์์ฐ์ค๋ฝ๊ณ ๋ฐ๋ปํ๋ฉฐ, ์์์์ ์ค๋ ๋ฏํ ๋ถ๋๋ฌ์ด ๋น์ด ์ฅ๋ฉด์ ๋น์ถฅ๋๋ค. ์ฅ๋ฉด์ ์ค์ ์์์ฒ๋ผ ๋ณด์
๋๋ค.",
|
| 552 |
+
lines=5,
|
| 553 |
+
)
|
| 554 |
+
img2vid_enhance_toggle = Toggle(
|
| 555 |
+
label="ํ๋กฌํํธ ๊ฐ์ ",
|
| 556 |
+
value=False,
|
| 557 |
+
interactive=True,
|
| 558 |
+
)
|
| 559 |
+
img2vid_negative_prompt = gr.Textbox(
|
| 560 |
+
label="Step 3: ๋ค๊ฑฐํฐ๋ธ ํ๋กฌํํธ ์
๋ ฅ",
|
| 561 |
+
placeholder="๋น๋์ค์์ ์ํ์ง ์๋ ์์๋ฅผ ์ค๋ช
ํ์ธ์...",
|
| 562 |
+
value="low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 563 |
+
lines=2,
|
| 564 |
+
)
|
| 565 |
+
|
| 566 |
+
img2vid_preset = gr.Dropdown(
|
| 567 |
+
choices=[p["label"] for p in preset_options],
|
| 568 |
+
value="768x512, 97 frames",
|
| 569 |
+
label="Step 3.1: ํด์๋ ํ๋ฆฌ์
์ ํ",
|
| 570 |
+
)
|
| 571 |
+
|
| 572 |
+
img2vid_frame_rate = gr.Slider(
|
| 573 |
+
label="Step 3.2: ํ๋ ์ ๋ ์ดํธ",
|
| 574 |
+
minimum=21,
|
| 575 |
+
maximum=30,
|
| 576 |
+
step=1,
|
| 577 |
+
value=25,
|
| 578 |
+
)
|
| 579 |
+
|
| 580 |
+
img2vid_advanced = create_advanced_options()
|
| 581 |
+
img2vid_generate = gr.Button(
|
| 582 |
+
"Step 6: ๋น๋์ค ์์ฑ", variant="primary", size="lg"
|
| 583 |
+
)
|
| 584 |
+
|
| 585 |
+
with gr.Column():
|
| 586 |
+
img2vid_output = gr.Video(label="์์ฑ๋ ๋น๋์ค")
|
| 587 |
+
|
| 588 |
+
with gr.Row():
|
| 589 |
+
gr.Examples(
|
| 590 |
+
examples=[
|
| 591 |
+
[
|
| 592 |
+
"assets/i2v_i2.png",
|
| 593 |
+
"์ฌ์ฑ์ด ํฐ์ ์ ๊ธฐ ๋ฒ๋ ์์์ ๋๋ ๋ฌผ์ด ๋ด๊ธด ๋๋น๋ฅผ ์ ๊ณ ์์ต๋๋ค. ๋ณด๋ผ์ ๋งค๋ํ์ด๋ฅผ ๋ฐ๋ฅธ ๊ทธ๋
์ ์์ด ํ์ ๋๋น ์์์ ๋๋ฌด ์๊ฐ๋ฝ์ ์ํ์ผ๋ก ์์ง์
๋๋ค. ๋๋น๋ ๊ฒ์์ ๋ฒํผ๊ณผ ๋์งํธ ๋์คํ๋ ์ด๊ฐ ์๋ ํฐ์ ์ ๊ธฐ ๋ฒ๋ ์์ ๋์ฌ ์์ต๋๋ค. ๋ฒ๋๋ ์ค๋ฅธ์ชฝ ์๋ ๋ชจ์๋ฆฌ์ ๋นจ๊ฐ์๊ณผ ํฐ์ ์ฒดํฌ๋ฌด๋ฌ ์ฒ์ด ๋ถ๋ถ์ ์ผ๋ก ๋ณด์ด๋ ํฐ์ ์กฐ๋ฆฌ๋ ์์ ๋์ฌ ์์ต๋๋ค. ์นด๋ฉ๋ผ ๊ฐ๋๋ ์ ํํ ์์์ ๋ด๋ ค๋ค๋ณด๋ ๊ฐ๋์ด๋ฉฐ ์ฅ๋ฉด ๋ด๋ด ๊ณ ์ ๋์ด ์์ต๋๋ค. ์กฐ๋ช
์ ๋ฐ๊ณ ๊ณ ๋ฅธ ์ค์ฑ์ ์ธ ํฐ์ ๋น์ผ๋ก ์ฅ๋ฉด์ ๋น์ถฅ๋๋ค. ์ฅ๋ฉด์ ์ค์ ์์์ฒ๋ผ ๋ณด์
๋๋ค.",
|
| 594 |
+
"low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 595 |
+
"assets/i2v_2.mp4",
|
| 596 |
+
],
|
| 597 |
+
[
|
| 598 |
+
"assets/i2v_i0.png",
|
| 599 |
+
"๊ธด ํ๋ฅด๋ ๋๋ ์ค๋ฅผ ์
์ ์ฌ์ฑ์ด ๋คํ์ ์์ ๋ฑ์ ์นด๋ฉ๋ผ๋ฅผ ํฅํ ์ฑ ์งํ์ ์ ๋ฐ๋ผ๋ณด๊ณ ์์ต๋๋ค. ๊ทธ๋
์ ๋จธ๋ฆฌ์นด๋ฝ์ ๊ธธ๊ณ ๋ฐ์ผ๋ฉฐ ๋ฑ ์๋๋ก ํ๋ฌ๋ด๋ฆฝ๋๋ค. ๊ทธ๋
๋ ํฐ ์ฐธ๋๋ฌด์ ๋๊ฒ ํผ์ง ๊ฐ์ง ์๋์ ์ ์์ต๋๋ค. ์ผ์ชฝ์ผ๋ก๋ ๋ง๋ผ๋ถ์ ์๋ ์์ ํด๋์ํ ๋ฏธ๊ตญ ์๋์ฐจ๊ฐ ์ฃผ์ฐจ๋์ด ์์ต๋๋ค. ๋ฉ๋ฆฌ์๋ ํ ๋์ ๋ถ์์ง ์๋์ฐจ๊ฐ ์์ผ๋ก ๋์ ์์ต๋๋ค. ์์ ํ๋์ ์ด๋์ด ํ๋์ ๋ฐฐ๊ฒฝ์ผ๋ก ๋ฐ์ ํฐ ๊ตฌ๋ฆ์ด ๊ทน์ ์ธ ์บ๋ฒ์ค๋ฅผ ์ด๋ฃจ๊ณ ์์ต๋๋ค. ์ ์ฒด ์ด๋ฏธ์ง๋ ํ๋ฐฑ์ผ๋ก, ๋น๊ณผ ๊ทธ๋ฆผ์์ ๋๋น๋ฅผ ๊ฐ์กฐํฉ๋๋ค. ์ฌ์ฑ์ด ์ฒ์ฒํ ์๋์ฐจ๋ฅผ ํฅํด ๊ฑธ์ด๊ฐ๊ณ ์์ต๋๋ค.",
|
| 600 |
+
"low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 601 |
+
"assets/i2v_0.mp4",
|
| 602 |
+
],
|
| 603 |
+
[
|
| 604 |
+
"assets/i2v_i1.png",
|
| 605 |
+
"ํ ์์ ์์ด ๋์๊ธฐ ๋ฌผ๋ ์์์ ์ ํ ์กฐ๊ฐ์ ๋ชจ์ ์ก์ ์ ์ฐจ์ ์ผ๋ก ์๋ฟ ๋ชจ์์ ๋ง๋ค์ด๊ฐ๊ณ ์์ต๋๋ค. ํ๋ ์ ๋ฐ์ ์ฌ๋์ ์์ด ์ ํ ๋ก ๋ฎ์ฌ ์์ผ๋ฉฐ, ํ์ ํ๋ ๋์๊ธฐ ๋ฌผ๋ ์ค์์ ์ ํ ๋ฉ์ด๋ฆฌ๋ฅผ ๋ถ๋๋ฝ๊ฒ ๋๋ฅด๊ณ ์์ต๋๋ค. ์์ ์ํ์ผ๋ก ์์ง์ด๋ฉฐ, ์ ํ ์์ชฝ์ ์ ์ฐจ์ ์ผ๋ก ์๋ฟ ๋ชจ์์ ๋ง๋ค์ด๊ฐ๋๋ค. ์นด๋ฉ๋ผ๋ ๋์๊ธฐ ๋ฌผ๋ ๋ฐ๋ก ์์ ์์นํ์ฌ ์ ํ ๊ฐ ๋ชจ์ ์กํ๊ฐ๋ ๊ฒ์ ์กฐ๊ฐ๋๋ก ๋ณด์ฌ์ค๋๋ค. ์กฐ๋ช
์ ๋ฐ๊ณ ๊ณ ๋ฅด๋ฉฐ, ์ ํ ์ ๊ทธ๊ฒ์ ๋ค๋ฃจ๋ ์์ ๋ฐ๊ฒ ๋น์ถฅ๋๋ค. ์ฅ๋ฉด์ ์ค์ ์์์ฒ๋ผ ์ดฌ์๋์์ต๋๋ค.",
|
| 606 |
+
"low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive",
|
| 607 |
+
"assets/i2v_1.mp4",
|
| 608 |
+
],
|
| 609 |
+
],
|
| 610 |
+
inputs=[
|
| 611 |
+
img2vid_image,
|
| 612 |
+
img2vid_prompt,
|
| 613 |
+
img2vid_negative_prompt,
|
| 614 |
+
img2vid_output,
|
| 615 |
+
],
|
| 616 |
+
label="์ด๋ฏธ์ง-๋น๋์ค ์์ฑ ์์",
|
| 617 |
+
)
|
| 618 |
+
|
| 619 |
+
# Event handlers
|
| 620 |
+
# Event handlers
|
| 621 |
+
txt2vid_preset.change(
|
| 622 |
+
fn=preset_changed,
|
| 623 |
+
inputs=[txt2vid_preset],
|
| 624 |
+
outputs=txt2vid_advanced[3:]
|
| 625 |
+
)
|
| 626 |
+
|
| 627 |
+
txt2vid_enhance_toggle.change(
|
| 628 |
+
fn=update_prompt_t2v,
|
| 629 |
+
inputs=[txt2vid_prompt, txt2vid_enhance_toggle],
|
| 630 |
+
outputs=txt2vid_prompt
|
| 631 |
+
)
|
| 632 |
+
|
| 633 |
+
txt2vid_generate.click(
|
| 634 |
+
fn=generate_video_from_text,
|
| 635 |
+
inputs=[
|
| 636 |
+
txt2vid_prompt,
|
| 637 |
+
txt2vid_enhance_toggle,
|
| 638 |
+
txt2vid_negative_prompt,
|
| 639 |
+
txt2vid_frame_rate,
|
| 640 |
+
*txt2vid_advanced,
|
| 641 |
+
],
|
| 642 |
+
outputs=txt2vid_output,
|
| 643 |
+
concurrency_limit=1,
|
| 644 |
+
concurrency_id="generate_video",
|
| 645 |
+
queue=True,
|
| 646 |
+
)
|
| 647 |
+
|
| 648 |
+
img2vid_preset.change(
|
| 649 |
+
fn=preset_changed,
|
| 650 |
+
inputs=[img2vid_preset],
|
| 651 |
+
outputs=img2vid_advanced[3:]
|
| 652 |
+
)
|
| 653 |
+
|
| 654 |
+
img2vid_enhance_toggle.change(
|
| 655 |
+
fn=update_prompt_i2v,
|
| 656 |
+
inputs=[img2vid_prompt, img2vid_enhance_toggle],
|
| 657 |
+
outputs=img2vid_prompt
|
| 658 |
+
)
|
| 659 |
+
|
| 660 |
+
img2vid_generate.click(
|
| 661 |
+
fn=generate_video_from_image,
|
| 662 |
+
inputs=[
|
| 663 |
+
img2vid_image,
|
| 664 |
+
img2vid_prompt,
|
| 665 |
+
img2vid_enhance_toggle,
|
| 666 |
+
img2vid_negative_prompt,
|
| 667 |
+
img2vid_frame_rate,
|
| 668 |
+
*img2vid_advanced,
|
| 669 |
+
],
|
| 670 |
+
outputs=img2vid_output,
|
| 671 |
+
concurrency_limit=1,
|
| 672 |
+
concurrency_id="generate_video",
|
| 673 |
+
queue=True,
|
| 674 |
+
)
|
| 675 |
+
|
| 676 |
+
if __name__ == "__main__":
|
| 677 |
+
iface.queue(max_size=64, default_concurrency_limit=1, api_open=False).launch(
|
| 678 |
+
share=True, show_api=False
|
| 679 |
+
)
|