|
from diffusers_helper.hf_login import login |
|
|
|
import os |
|
import threading |
|
import time |
|
import requests |
|
from requests.adapters import HTTPAdapter |
|
from urllib3.util.retry import Retry |
|
import json |
|
|
|
os.environ['HF_HOME'] = os.path.abspath(os.path.realpath(os.path.join(os.path.dirname(__file__), './hf_download'))) |
|
|
|
import gradio as gr |
|
import torch |
|
import traceback |
|
import einops |
|
import safetensors.torch as sf |
|
import numpy as np |
|
import math |
|
|
|
|
|
IN_HF_SPACE = os.environ.get('SPACE_ID') is not None |
|
|
|
|
|
GPU_AVAILABLE = False |
|
GPU_INITIALIZED = False |
|
last_update_time = time.time() |
|
|
|
|
|
if IN_HF_SPACE: |
|
try: |
|
import spaces |
|
print("Hugging Face Space環境内で実行中、spacesモジュールをインポートしました") |
|
|
|
|
|
try: |
|
GPU_AVAILABLE = torch.cuda.is_available() |
|
print(f"GPU利用可能: {GPU_AVAILABLE}") |
|
if GPU_AVAILABLE: |
|
print(f"GPUデバイス名: {torch.cuda.get_device_name(0)}") |
|
print(f"GPUメモリ: {torch.cuda.get_device_properties(0).total_memory / 1e9} GB") |
|
|
|
|
|
test_tensor = torch.zeros(1, device='cuda') |
|
test_tensor = test_tensor + 1 |
|
del test_tensor |
|
print("GPUテスト操作に成功しました") |
|
else: |
|
print("警告: CUDAが利用可能と報告されていますが、GPUデバイスが検出されませんでした") |
|
except Exception as e: |
|
GPU_AVAILABLE = False |
|
print(f"GPU確認中にエラーが発生しました: {e}") |
|
print("CPUモードで実行します") |
|
except ImportError: |
|
print("spacesモジュールのインポートに失敗しました。Hugging Face Space環境外かもしれません") |
|
GPU_AVAILABLE = torch.cuda.is_available() |
|
|
|
from PIL import Image |
|
from diffusers import AutoencoderKLHunyuanVideo |
|
from transformers import LlamaModel, CLIPTextModel, LlamaTokenizerFast, CLIPTokenizer |
|
from diffusers_helper.hunyuan import encode_prompt_conds, vae_decode, vae_encode, vae_decode_fake |
|
from diffusers_helper.utils import save_bcthw_as_mp4, crop_or_pad_yield_mask, soft_append_bcthw, resize_and_center_crop, state_dict_weighted_merge, state_dict_offset_merge, generate_timestamp |
|
from diffusers_helper.models.hunyuan_video_packed import HunyuanVideoTransformer3DModelPacked |
|
from diffusers_helper.pipelines.k_diffusion_hunyuan import sample_hunyuan |
|
from diffusers_helper.memory import cpu, gpu, get_cuda_free_memory_gb, move_model_to_device_with_memory_preservation, offload_model_from_device_for_memory_preservation, fake_diffusers_current_device, DynamicSwapInstaller, unload_complete_models, load_model_as_complete, IN_HF_SPACE as MEMORY_IN_HF_SPACE |
|
from diffusers_helper.thread_utils import AsyncStream, async_run |
|
from diffusers_helper.gradio.progress_bar import make_progress_bar_css, make_progress_bar_html |
|
from transformers import SiglipImageProcessor, SiglipVisionModel |
|
from diffusers_helper.clip_vision import hf_clip_vision_encode |
|
from diffusers_helper.bucket_tools import find_nearest_bucket |
|
|
|
outputs_folder = './outputs/' |
|
os.makedirs(outputs_folder, exist_ok=True) |
|
|
|
|
|
if not IN_HF_SPACE: |
|
|
|
try: |
|
if torch.cuda.is_available(): |
|
free_mem_gb = get_cuda_free_memory_gb(gpu) |
|
print(f'空きVRAM {free_mem_gb} GB') |
|
else: |
|
free_mem_gb = 6.0 |
|
print("CUDAが利用できません。デフォルトのメモリ設定を使用します") |
|
except Exception as e: |
|
free_mem_gb = 6.0 |
|
print(f"CUDAメモリ取得中にエラーが発生しました: {e}、デフォルトのメモリ設定を使用します") |
|
|
|
high_vram = free_mem_gb > 60 |
|
print(f'高VRAM モード: {high_vram}') |
|
else: |
|
|
|
print("Spaces環境でデフォルトのメモリ設定を使用します") |
|
try: |
|
if GPU_AVAILABLE: |
|
free_mem_gb = torch.cuda.get_device_properties(0).total_memory / 1e9 * 0.9 |
|
high_vram = free_mem_gb > 10 |
|
else: |
|
free_mem_gb = 6.0 |
|
high_vram = False |
|
except Exception as e: |
|
print(f"GPUメモリ取得中にエラーが発生しました: {e}") |
|
free_mem_gb = 6.0 |
|
high_vram = False |
|
|
|
print(f'GPUメモリ: {free_mem_gb:.2f} GB, 高VRAMモード: {high_vram}') |
|
|
|
|
|
models = {} |
|
cpu_fallback_mode = not GPU_AVAILABLE |
|
|
|
|
|
def load_models(): |
|
global models, cpu_fallback_mode, GPU_INITIALIZED |
|
|
|
if GPU_INITIALIZED: |
|
print("モデルはすでに読み込まれています。重複読み込みをスキップします") |
|
return models |
|
|
|
print("モデルの読み込みを開始しています...") |
|
|
|
try: |
|
|
|
device = 'cuda' if GPU_AVAILABLE and not cpu_fallback_mode else 'cpu' |
|
model_device = 'cpu' |
|
|
|
|
|
dtype = torch.float16 if GPU_AVAILABLE else torch.float32 |
|
transformer_dtype = torch.bfloat16 if GPU_AVAILABLE else torch.float32 |
|
|
|
print(f"使用デバイス: {device}, モデル精度: {dtype}, Transformer精度: {transformer_dtype}") |
|
|
|
|
|
try: |
|
text_encoder = LlamaModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder', torch_dtype=dtype).to(model_device) |
|
text_encoder_2 = CLIPTextModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder_2', torch_dtype=dtype).to(model_device) |
|
tokenizer = LlamaTokenizerFast.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer') |
|
tokenizer_2 = CLIPTokenizer.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer_2') |
|
vae = AutoencoderKLHunyuanVideo.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='vae', torch_dtype=dtype).to(model_device) |
|
|
|
feature_extractor = SiglipImageProcessor.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='feature_extractor') |
|
image_encoder = SiglipVisionModel.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='image_encoder', torch_dtype=dtype).to(model_device) |
|
|
|
transformer = HunyuanVideoTransformer3DModelPacked.from_pretrained('tori29umai/FramePackI2V_HY_rotate_landscape', torch_dtype=transformer_dtype).to(model_device) |
|
|
|
print("すべてのモデルの読み込みに成功しました") |
|
except Exception as e: |
|
print(f"モデル読み込み中にエラーが発生しました: {e}") |
|
print("精度を下げて再試行します...") |
|
|
|
|
|
dtype = torch.float32 |
|
transformer_dtype = torch.float32 |
|
cpu_fallback_mode = True |
|
|
|
text_encoder = LlamaModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder', torch_dtype=dtype).to('cpu') |
|
text_encoder_2 = CLIPTextModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder_2', torch_dtype=dtype).to('cpu') |
|
tokenizer = LlamaTokenizerFast.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer') |
|
tokenizer_2 = CLIPTokenizer.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer_2') |
|
vae = AutoencoderKLHunyuanVideo.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='vae', torch_dtype=dtype).to('cpu') |
|
|
|
feature_extractor = SiglipImageProcessor.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='feature_extractor') |
|
image_encoder = SiglipVisionModel.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='image_encoder', torch_dtype=dtype).to('cpu') |
|
|
|
transformer = HunyuanVideoTransformer3DModelPacked.from_pretrained('tori29umai/FramePackI2V_HY_rotate_landscape', torch_dtype=transformer_dtype).to('cpu') |
|
|
|
print("CPUモードですべてのモデルの読み込みに成功しました") |
|
|
|
vae.eval() |
|
text_encoder.eval() |
|
text_encoder_2.eval() |
|
image_encoder.eval() |
|
transformer.eval() |
|
|
|
if not high_vram or cpu_fallback_mode: |
|
vae.enable_slicing() |
|
vae.enable_tiling() |
|
|
|
transformer.high_quality_fp32_output_for_inference = True |
|
print('transformer.high_quality_fp32_output_for_inference = True') |
|
|
|
|
|
if not cpu_fallback_mode: |
|
transformer.to(dtype=transformer_dtype) |
|
vae.to(dtype=dtype) |
|
image_encoder.to(dtype=dtype) |
|
text_encoder.to(dtype=dtype) |
|
text_encoder_2.to(dtype=dtype) |
|
|
|
vae.requires_grad_(False) |
|
text_encoder.requires_grad_(False) |
|
text_encoder_2.requires_grad_(False) |
|
image_encoder.requires_grad_(False) |
|
transformer.requires_grad_(False) |
|
|
|
if torch.cuda.is_available() and not cpu_fallback_mode: |
|
try: |
|
if not high_vram: |
|
|
|
DynamicSwapInstaller.install_model(transformer, device=device) |
|
DynamicSwapInstaller.install_model(text_encoder, device=device) |
|
else: |
|
text_encoder.to(device) |
|
text_encoder_2.to(device) |
|
image_encoder.to(device) |
|
vae.to(device) |
|
transformer.to(device) |
|
print(f"モデルを{device}デバイスに移動することに成功しました") |
|
except Exception as e: |
|
print(f"モデルを{device}に移動中にエラーが発生しました: {e}") |
|
print("CPUモードにフォールバックします") |
|
cpu_fallback_mode = True |
|
|
|
|
|
models = { |
|
'text_encoder': text_encoder, |
|
'text_encoder_2': text_encoder_2, |
|
'tokenizer': tokenizer, |
|
'tokenizer_2': tokenizer_2, |
|
'vae': vae, |
|
'feature_extractor': feature_extractor, |
|
'image_encoder': image_encoder, |
|
'transformer': transformer |
|
} |
|
|
|
GPU_INITIALIZED = True |
|
print(f"モデルの読み込みが完了しました。実行モード: {'CPU' if cpu_fallback_mode else 'GPU'}") |
|
return models |
|
except Exception as e: |
|
print(f"モデル読み込みプロセスでエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
|
|
|
|
error_info = { |
|
"error": str(e), |
|
"traceback": traceback.format_exc(), |
|
"cuda_available": torch.cuda.is_available(), |
|
"device": "cpu" if cpu_fallback_mode else "cuda", |
|
} |
|
|
|
|
|
try: |
|
with open(os.path.join(outputs_folder, "error_log.txt"), "w") as f: |
|
f.write(str(error_info)) |
|
except: |
|
pass |
|
|
|
|
|
cpu_fallback_mode = True |
|
return {} |
|
|
|
|
|
|
|
if IN_HF_SPACE and 'spaces' in globals() and GPU_AVAILABLE: |
|
try: |
|
@spaces.GPU |
|
def initialize_models(): |
|
"""@spaces.GPU装飾子内でモデルを初期化""" |
|
global GPU_INITIALIZED |
|
try: |
|
result = load_models() |
|
GPU_INITIALIZED = True |
|
return result |
|
except Exception as e: |
|
print(f"spaces.GPUを使用したモデル初期化中にエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
global cpu_fallback_mode |
|
cpu_fallback_mode = True |
|
|
|
return load_models() |
|
except Exception as e: |
|
print(f"spaces.GPU装飾子の作成中にエラーが発生しました: {e}") |
|
|
|
def initialize_models(): |
|
return load_models() |
|
|
|
|
|
|
|
def get_models(): |
|
"""モデルを取得し、まだ読み込まれていない場合は読み込む""" |
|
global models, GPU_INITIALIZED |
|
|
|
|
|
model_loading_key = "__model_loading__" |
|
|
|
if not models: |
|
|
|
if model_loading_key in globals(): |
|
print("モデルは現在読み込み中です。お待ちください...") |
|
|
|
import time |
|
start_wait = time.time() |
|
while not models and model_loading_key in globals(): |
|
time.sleep(0.5) |
|
|
|
if time.time() - start_wait > 60: |
|
print("モデル読み込み待機がタイムアウトしました") |
|
break |
|
|
|
if models: |
|
return models |
|
|
|
try: |
|
|
|
globals()[model_loading_key] = True |
|
|
|
if IN_HF_SPACE and 'spaces' in globals() and GPU_AVAILABLE and not cpu_fallback_mode: |
|
try: |
|
print("@spaces.GPU装飾子を使用してモデルを読み込みます") |
|
models = initialize_models() |
|
except Exception as e: |
|
print(f"GPU装飾子を使用したモデル読み込みに失敗しました: {e}") |
|
print("直接モデルを読み込みます") |
|
models = load_models() |
|
else: |
|
print("モデルを直接読み込みます") |
|
models = load_models() |
|
except Exception as e: |
|
print(f"モデル読み込み中に予期しないエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
|
|
models = {} |
|
finally: |
|
|
|
if model_loading_key in globals(): |
|
del globals()[model_loading_key] |
|
|
|
return models |
|
|
|
|
|
|
|
PREDEFINED_RESOLUTIONS = [ |
|
(416, 960), (448, 864), (480, 832), (512, 768), (544, 704), |
|
(576, 672), (608, 640), (640, 608), (672, 576), (704, 544), |
|
(768, 512), (832, 480), (864, 448), (960, 416) |
|
] |
|
|
|
|
|
def find_closest_aspect_ratio(width, height, target_resolutions): |
|
""" |
|
事前定義された解像度リストから、元の画像のアスペクト比に最も近い解像度を見つける |
|
|
|
引数: |
|
width: 元の画像の幅 |
|
height: 元の画像の高さ |
|
target_resolutions: 目標解像度のリスト(幅, 高さ)のタプル |
|
|
|
戻り値: |
|
tuple: 最も近いアスペクト比の (target_width, target_height) |
|
""" |
|
original_aspect = width / height |
|
|
|
|
|
min_diff = float('inf') |
|
closest_resolution = None |
|
|
|
for target_width, target_height in target_resolutions: |
|
target_aspect = target_width / target_height |
|
diff = abs(original_aspect - target_aspect) |
|
|
|
if diff < min_diff: |
|
min_diff = diff |
|
closest_resolution = (target_width, target_height) |
|
|
|
return closest_resolution |
|
|
|
|
|
stream = AsyncStream() |
|
|
|
|
|
@torch.no_grad() |
|
def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache): |
|
global last_update_time |
|
last_update_time = time.time() |
|
|
|
|
|
total_second_length = min(total_second_length, 3.0) |
|
|
|
|
|
try: |
|
models = get_models() |
|
if not models: |
|
error_msg = "モデルの読み込みに失敗しました。詳細情報はログを確認してください" |
|
print(error_msg) |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
text_encoder = models['text_encoder'] |
|
text_encoder_2 = models['text_encoder_2'] |
|
tokenizer = models['tokenizer'] |
|
tokenizer_2 = models['tokenizer_2'] |
|
vae = models['vae'] |
|
feature_extractor = models['feature_extractor'] |
|
image_encoder = models['image_encoder'] |
|
transformer = models['transformer'] |
|
except Exception as e: |
|
error_msg = f"モデル取得中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
device = 'cuda' if GPU_AVAILABLE and not cpu_fallback_mode else 'cpu' |
|
print(f"推論に使用するデバイス: {device}") |
|
|
|
|
|
if cpu_fallback_mode: |
|
print("CPUモードではより軽量なパラメータを使用します") |
|
|
|
latent_window_size = min(latent_window_size, 5) |
|
steps = min(steps, 15) |
|
total_second_length = min(total_second_length, 2.0) |
|
|
|
total_latent_sections = (total_second_length * 30) / (latent_window_size * 4) |
|
total_latent_sections = int(max(round(total_latent_sections), 1)) |
|
|
|
job_id = generate_timestamp() |
|
last_output_filename = None |
|
history_pixels = None |
|
history_latents = None |
|
total_generated_latent_frames = 0 |
|
|
|
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, '開始中 ...')))) |
|
|
|
try: |
|
|
|
if not high_vram and not cpu_fallback_mode: |
|
try: |
|
unload_complete_models( |
|
text_encoder, text_encoder_2, image_encoder, vae, transformer |
|
) |
|
except Exception as e: |
|
print(f"モデルのアンロード中にエラーが発生しました: {e}") |
|
|
|
|
|
|
|
last_update_time = time.time() |
|
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'テキストエンコーディング中 ...')))) |
|
|
|
try: |
|
if not high_vram and not cpu_fallback_mode: |
|
fake_diffusers_current_device(text_encoder, device) |
|
load_model_as_complete(text_encoder_2, target_device=device) |
|
|
|
llama_vec, clip_l_pooler = encode_prompt_conds(prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2) |
|
|
|
if cfg == 1: |
|
llama_vec_n, clip_l_pooler_n = torch.zeros_like(llama_vec), torch.zeros_like(clip_l_pooler) |
|
else: |
|
llama_vec_n, clip_l_pooler_n = encode_prompt_conds(n_prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2) |
|
|
|
llama_vec, llama_attention_mask = crop_or_pad_yield_mask(llama_vec, length=512) |
|
llama_vec_n, llama_attention_mask_n = crop_or_pad_yield_mask(llama_vec_n, length=512) |
|
except Exception as e: |
|
error_msg = f"テキストエンコーディング中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
try: |
|
H, W, C = input_image.shape |
|
|
|
|
|
target_width, target_height = find_closest_aspect_ratio(W, H, PREDEFINED_RESOLUTIONS) |
|
|
|
|
|
width = target_width |
|
height = target_height |
|
|
|
|
|
if cpu_fallback_mode: |
|
scale_factor = min(320 / target_height, 320 / target_width) |
|
target_height = int(target_height * scale_factor) |
|
target_width = int(target_width * scale_factor) |
|
|
|
height = target_height |
|
width = target_width |
|
|
|
print(f'元の画像サイズ: {W}x{H}, リサイズ先: {target_width}x{target_height}') |
|
|
|
|
|
input_image_np = resize_and_center_crop(input_image, target_width=target_width, target_height=target_height) |
|
Image.fromarray(input_image_np).save(os.path.join(outputs_folder, f'{job_id}.png')) |
|
|
|
input_image_pt = torch.from_numpy(input_image_np).float() / 127.5 - 1 |
|
input_image_pt = input_image_pt.permute(2, 0, 1)[None, :, None] |
|
except Exception as e: |
|
error_msg = f"画像処理中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
last_update_time = time.time() |
|
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'VAEエンコーディング中 ...')))) |
|
|
|
try: |
|
if not high_vram and not cpu_fallback_mode: |
|
load_model_as_complete(vae, target_device=device) |
|
|
|
start_latent = vae_encode(input_image_pt, vae) |
|
except Exception as e: |
|
error_msg = f"VAEエンコーディング中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
last_update_time = time.time() |
|
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'CLIP Visionエンコーディング中 ...')))) |
|
|
|
try: |
|
if not high_vram and not cpu_fallback_mode: |
|
load_model_as_complete(image_encoder, target_device=device) |
|
|
|
image_encoder_output = hf_clip_vision_encode(input_image_np, feature_extractor, image_encoder) |
|
image_encoder_last_hidden_state = image_encoder_output.last_hidden_state |
|
except Exception as e: |
|
error_msg = f"CLIP Visionエンコーディング中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
try: |
|
llama_vec = llama_vec.to(transformer.dtype) |
|
llama_vec_n = llama_vec_n.to(transformer.dtype) |
|
clip_l_pooler = clip_l_pooler.to(transformer.dtype) |
|
clip_l_pooler_n = clip_l_pooler_n.to(transformer.dtype) |
|
image_encoder_last_hidden_state = image_encoder_last_hidden_state.to(transformer.dtype) |
|
except Exception as e: |
|
error_msg = f"データ型変換中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
last_update_time = time.time() |
|
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'サンプリング開始 ...')))) |
|
|
|
rnd = torch.Generator("cpu").manual_seed(seed) |
|
num_frames = latent_window_size * 4 - 3 |
|
|
|
try: |
|
history_latents = torch.zeros(size=(1, 16, 1 + 2 + 16, height // 8, width // 8), dtype=torch.float32).cpu() |
|
history_pixels = None |
|
total_generated_latent_frames = 0 |
|
except Exception as e: |
|
error_msg = f"履歴状態の初期化中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
latent_paddings = reversed(range(total_latent_sections)) |
|
|
|
if total_latent_sections > 4: |
|
|
|
|
|
|
|
|
|
|
|
latent_paddings = [3] + [2] * (total_latent_sections - 3) + [1, 0] |
|
|
|
for latent_padding in latent_paddings: |
|
last_update_time = time.time() |
|
is_last_section = latent_padding == 0 |
|
latent_padding_size = latent_padding * latent_window_size |
|
|
|
if stream.input_queue.top() == 'end': |
|
|
|
if history_pixels is not None and total_generated_latent_frames > 0: |
|
try: |
|
output_filename = os.path.join(outputs_folder, f'{job_id}_final_{total_generated_latent_frames}.mp4') |
|
save_bcthw_as_mp4(history_pixels, output_filename, fps=30, crf=18) |
|
stream.output_queue.push(('file', output_filename)) |
|
except Exception as e: |
|
print(f"最終動画保存中にエラーが発生しました: {e}") |
|
|
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
print(f'latent_padding_size = {latent_padding_size}, is_last_section = {is_last_section}') |
|
|
|
try: |
|
indices = torch.arange(0, sum([1, latent_padding_size, latent_window_size, 1, 2, 16])).unsqueeze(0) |
|
clean_latent_indices_pre, blank_indices, latent_indices, clean_latent_indices_post, clean_latent_2x_indices, clean_latent_4x_indices = indices.split([1, latent_padding_size, latent_window_size, 1, 2, 16], dim=1) |
|
clean_latent_indices = torch.cat([clean_latent_indices_pre, clean_latent_indices_post], dim=1) |
|
|
|
clean_latents_pre = start_latent.to(history_latents) |
|
clean_latents_post, clean_latents_2x, clean_latents_4x = history_latents[:, :, :1 + 2 + 16, :, :].split([1, 2, 16], dim=2) |
|
clean_latents = torch.cat([clean_latents_pre, clean_latents_post], dim=2) |
|
except Exception as e: |
|
error_msg = f"サンプリングデータ準備中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
|
|
if last_output_filename: |
|
stream.output_queue.push(('file', last_output_filename)) |
|
continue |
|
|
|
if not high_vram and not cpu_fallback_mode: |
|
try: |
|
unload_complete_models() |
|
move_model_to_device_with_memory_preservation(transformer, target_device=device, preserved_memory_gb=gpu_memory_preservation) |
|
except Exception as e: |
|
print(f"transformerをGPUに移動中にエラーが発生しました: {e}") |
|
|
|
|
|
if use_teacache and not cpu_fallback_mode: |
|
try: |
|
transformer.initialize_teacache(enable_teacache=True, num_steps=steps) |
|
except Exception as e: |
|
print(f"teacache初期化中にエラーが発生しました: {e}") |
|
|
|
transformer.initialize_teacache(enable_teacache=False) |
|
else: |
|
transformer.initialize_teacache(enable_teacache=False) |
|
|
|
def callback(d): |
|
global last_update_time |
|
last_update_time = time.time() |
|
|
|
try: |
|
|
|
print(f"【デバッグ】コールバック関数: ステップ {d['i']}, 停止信号のチェック") |
|
try: |
|
queue_top = stream.input_queue.top() |
|
print(f"【デバッグ】コールバック関数: キュー先頭信号 = {queue_top}") |
|
|
|
if queue_top == 'end': |
|
print("【デバッグ】コールバック関数: 停止信号を検出、中断準備中...") |
|
try: |
|
stream.output_queue.push(('end', None)) |
|
print("【デバッグ】コールバック関数: 出力キューにend信号を正常に送信") |
|
except Exception as e: |
|
print(f"【デバッグ】コールバック関数: 出力キューにend信号送信中にエラー: {e}") |
|
|
|
print("【デバッグ】コールバック関数: KeyboardInterrupt例外を投げる準備") |
|
raise KeyboardInterrupt('ユーザーによるタスク停止') |
|
except Exception as e: |
|
print(f"【デバッグ】コールバック関数: キュー先頭信号チェック中にエラー: {e}") |
|
|
|
preview = d['denoised'] |
|
preview = vae_decode_fake(preview) |
|
|
|
preview = (preview * 255.0).detach().cpu().numpy().clip(0, 255).astype(np.uint8) |
|
preview = einops.rearrange(preview, 'b c t h w -> (b h) (t w) c') |
|
|
|
current_step = d['i'] + 1 |
|
percentage = int(100.0 * current_step / steps) |
|
hint = f'サンプリング中 {current_step}/{steps}' |
|
desc = f'総生成フレーム数: {int(max(0, total_generated_latent_frames * 4 - 3))}, 動画長: {max(0, (total_generated_latent_frames * 4 - 3) / 30) :.2f} 秒 (FPS-30). 動画を現在拡張中...' |
|
stream.output_queue.push(('progress', (preview, desc, make_progress_bar_html(percentage, hint)))) |
|
except KeyboardInterrupt as e: |
|
|
|
print(f"【デバッグ】コールバック関数: KeyboardInterruptをキャッチ: {e}") |
|
print("【デバッグ】コールバック関数: 中断例外を再スロー、サンプリング関数に伝播") |
|
raise |
|
except Exception as e: |
|
print(f"【デバッグ】コールバック関数でエラー: {e}") |
|
|
|
print(f"【デバッグ】コールバック関数: ステップ {d['i']} 完了") |
|
return |
|
|
|
try: |
|
sampling_start_time = time.time() |
|
print(f"サンプリング開始、デバイス: {device}, データ型: {transformer.dtype}, TeaCache使用: {use_teacache and not cpu_fallback_mode}") |
|
|
|
try: |
|
print("【デバッグ】sample_hunyuanサンプリングプロセス開始") |
|
generated_latents = sample_hunyuan( |
|
transformer=transformer, |
|
sampler='unipc', |
|
width=width, |
|
height=height, |
|
frames=num_frames, |
|
real_guidance_scale=cfg, |
|
distilled_guidance_scale=gs, |
|
guidance_rescale=rs, |
|
|
|
num_inference_steps=steps, |
|
generator=rnd, |
|
prompt_embeds=llama_vec, |
|
prompt_embeds_mask=llama_attention_mask, |
|
prompt_poolers=clip_l_pooler, |
|
negative_prompt_embeds=llama_vec_n, |
|
negative_prompt_embeds_mask=llama_attention_mask_n, |
|
negative_prompt_poolers=clip_l_pooler_n, |
|
device=device, |
|
dtype=transformer.dtype, |
|
image_embeddings=image_encoder_last_hidden_state, |
|
latent_indices=latent_indices, |
|
clean_latents=clean_latents, |
|
clean_latent_indices=clean_latent_indices, |
|
clean_latents_2x=clean_latents_2x, |
|
clean_latent_2x_indices=clean_latent_2x_indices, |
|
clean_latents_4x=clean_latents_4x, |
|
clean_latent_4x_indices=clean_latent_4x_indices, |
|
callback=callback, |
|
) |
|
|
|
print(f"【デバッグ】サンプリング完了、所要時間: {time.time() - sampling_start_time:.2f}秒") |
|
except KeyboardInterrupt as e: |
|
|
|
print(f"【デバッグ】KeyboardInterruptをキャッチ: {e}") |
|
print("【デバッグ】ユーザーによるサンプリングプロセス中断、中断ロジック処理中") |
|
|
|
|
|
if last_output_filename: |
|
print(f"【デバッグ】部分的に生成された動画あり: {last_output_filename}、この動画を返します") |
|
stream.output_queue.push(('file', last_output_filename)) |
|
error_msg = "ユーザーにより生成プロセスが中断されましたが、部分的な動画は生成されています" |
|
else: |
|
print("【デバッグ】部分的に生成された動画なし、中断メッセージを返します") |
|
error_msg = "ユーザーにより生成プロセスが中断され、動画は生成されていません" |
|
|
|
print(f"【デバッグ】エラーメッセージを送信: {error_msg}") |
|
stream.output_queue.push(('error', error_msg)) |
|
print("【デバッグ】end信号を送信") |
|
stream.output_queue.push(('end', None)) |
|
print("【デバッグ】中断処理完了、リターン") |
|
return |
|
except Exception as e: |
|
print(f"サンプリングプロセス中にエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
|
|
|
|
if last_output_filename: |
|
stream.output_queue.push(('file', last_output_filename)) |
|
|
|
|
|
error_msg = f"サンプリングプロセス中にエラーが発生しましたが、部分的に生成された動画を返します: {e}" |
|
stream.output_queue.push(('error', error_msg)) |
|
else: |
|
|
|
error_msg = f"サンプリングプロセス中にエラーが発生し、動画を生成できませんでした: {e}" |
|
stream.output_queue.push(('error', error_msg)) |
|
|
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
try: |
|
if is_last_section: |
|
generated_latents = torch.cat([start_latent.to(generated_latents), generated_latents], dim=2) |
|
|
|
total_generated_latent_frames += int(generated_latents.shape[2]) |
|
history_latents = torch.cat([generated_latents.to(history_latents), history_latents], dim=2) |
|
except Exception as e: |
|
error_msg = f"生成された潜在変数の処理中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
traceback.print_exc() |
|
|
|
if last_output_filename: |
|
stream.output_queue.push(('file', last_output_filename)) |
|
stream.output_queue.push(('error', error_msg)) |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
if not high_vram and not cpu_fallback_mode: |
|
try: |
|
offload_model_from_device_for_memory_preservation(transformer, target_device=device, preserved_memory_gb=8) |
|
load_model_as_complete(vae, target_device=device) |
|
except Exception as e: |
|
print(f"モデルメモリ管理中にエラーが発生しました: {e}") |
|
|
|
|
|
try: |
|
real_history_latents = history_latents[:, :, :total_generated_latent_frames, :, :] |
|
except Exception as e: |
|
error_msg = f"履歴潜在変数の処理中にエラーが発生しました: {e}" |
|
print(error_msg) |
|
|
|
if last_output_filename: |
|
stream.output_queue.push(('file', last_output_filename)) |
|
continue |
|
|
|
try: |
|
vae_start_time = time.time() |
|
print(f"VAEデコード開始、潜在変数形状: {real_history_latents.shape}") |
|
|
|
if history_pixels is None: |
|
history_pixels = vae_decode(real_history_latents, vae).cpu() |
|
else: |
|
section_latent_frames = (latent_window_size * 2 + 1) if is_last_section else (latent_window_size * 2) |
|
overlapped_frames = latent_window_size * 4 - 3 |
|
|
|
current_pixels = vae_decode(real_history_latents[:, :, :section_latent_frames], vae).cpu() |
|
history_pixels = soft_append_bcthw(current_pixels, history_pixels, overlapped_frames) |
|
|
|
print(f"VAEデコード完了、所要時間: {time.time() - vae_start_time:.2f}秒") |
|
|
|
if not high_vram and not cpu_fallback_mode: |
|
try: |
|
unload_complete_models() |
|
except Exception as e: |
|
print(f"モデルのアンロード中にエラーが発生しました: {e}") |
|
|
|
output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4') |
|
|
|
save_start_time = time.time() |
|
save_bcthw_as_mp4(history_pixels, output_filename, fps=30, crf=18) |
|
print(f"動画保存完了、所要時間: {time.time() - save_start_time:.2f}秒") |
|
|
|
print(f'デコード完了。現在の潜在変数形状 {real_history_latents.shape}; ピクセル形状 {history_pixels.shape}') |
|
|
|
last_output_filename = output_filename |
|
stream.output_queue.push(('file', output_filename)) |
|
except Exception as e: |
|
print(f"動画のデコードまたは保存中にエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
|
|
|
|
if last_output_filename: |
|
stream.output_queue.push(('file', last_output_filename)) |
|
|
|
|
|
error_msg = f"動画のデコードまたは保存中にエラーが発生しました: {e}" |
|
stream.output_queue.push(('error', error_msg)) |
|
|
|
|
|
continue |
|
|
|
if is_last_section: |
|
break |
|
except Exception as e: |
|
print(f"【デバッグ】処理中にエラーが発生しました: {e}, タイプ: {type(e)}") |
|
print(f"【デバッグ】エラー詳細:") |
|
traceback.print_exc() |
|
|
|
|
|
if isinstance(e, KeyboardInterrupt): |
|
print("【デバッグ】外部KeyboardInterrupt例外を検出") |
|
|
|
if not high_vram and not cpu_fallback_mode: |
|
try: |
|
print("【デバッグ】リソース解放のためモデルをアンロード") |
|
unload_complete_models( |
|
text_encoder, text_encoder_2, image_encoder, vae, transformer |
|
) |
|
print("【デバッグ】モデルのアンロードに成功") |
|
except Exception as unload_error: |
|
print(f"【デバッグ】モデルのアンロード中にエラー: {unload_error}") |
|
pass |
|
|
|
|
|
if last_output_filename: |
|
print(f"【デバッグ】外部例外処理: 生成済み部分動画を返す {last_output_filename}") |
|
stream.output_queue.push(('file', last_output_filename)) |
|
else: |
|
print("【デバッグ】外部例外処理: 生成済み動画が見つかりません") |
|
|
|
|
|
error_msg = f"処理中にエラーが発生しました: {e}" |
|
print(f"【デバッグ】外部例外処理: エラーメッセージを送信: {error_msg}") |
|
stream.output_queue.push(('error', error_msg)) |
|
|
|
|
|
print("【デバッグ】ワーカー関数終了、end信号を送信") |
|
stream.output_queue.push(('end', None)) |
|
return |
|
|
|
|
|
|
|
if IN_HF_SPACE and 'spaces' in globals(): |
|
@spaces.GPU |
|
def process_with_gpu(input_image, prompt, n_prompt, seed, total_second_length, use_teacache): |
|
global stream |
|
assert input_image is not None, '入力画像がありません!' |
|
|
|
latent_window_size = 9 |
|
steps = 25 |
|
cfg = 1.0 |
|
gs = 10.0 |
|
rs = 0.0 |
|
gpu_memory_preservation = 6 |
|
|
|
|
|
|
|
yield None, None, '', '', gr.update(interactive=False), gr.update(interactive=True) |
|
|
|
try: |
|
stream = AsyncStream() |
|
|
|
|
|
async_run(worker, input_image, prompt, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache) |
|
|
|
output_filename = None |
|
prev_output_filename = None |
|
error_message = None |
|
|
|
|
|
while True: |
|
try: |
|
flag, data = stream.output_queue.next() |
|
|
|
if flag == 'file': |
|
output_filename = data |
|
prev_output_filename = output_filename |
|
|
|
yield output_filename, gr.update(), gr.update(), '', gr.update(interactive=False), gr.update(interactive=True) |
|
|
|
if flag == 'progress': |
|
preview, desc, html = data |
|
|
|
yield gr.update(), gr.update(visible=True, value=preview), desc, html, gr.update(interactive=False), gr.update(interactive=True) |
|
|
|
if flag == 'error': |
|
error_message = data |
|
print(f"エラーメッセージを受信: {error_message}") |
|
|
|
|
|
if flag == 'end': |
|
|
|
if output_filename is None and prev_output_filename is not None: |
|
output_filename = prev_output_filename |
|
|
|
|
|
if error_message: |
|
yield output_filename, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
else: |
|
|
|
yield output_filename, gr.update(visible=False), gr.update(), '', gr.update(interactive=True), gr.update(interactive=False) |
|
break |
|
except Exception as e: |
|
print(f"出力処理中にエラーが発生しました: {e}") |
|
|
|
current_time = time.time() |
|
if current_time - last_update_time > 60: |
|
print(f"処理がフリーズした可能性があります。{current_time - last_update_time:.1f}秒間更新がありません") |
|
|
|
|
|
if prev_output_filename: |
|
yield prev_output_filename, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
else: |
|
yield None, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
break |
|
|
|
except Exception as e: |
|
print(f"処理の開始中にエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
error_msg = str(e) |
|
|
|
yield None, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
|
|
process = process_with_gpu |
|
else: |
|
def process(input_image, prompt, n_prompt, seed, total_second_length, use_teacache): |
|
global stream |
|
assert input_image is not None, '入力画像がありません!' |
|
|
|
latent_window_size = 9 |
|
steps = 25 |
|
cfg = 1.0 |
|
gs = 10.0 |
|
rs = 0.0 |
|
gpu_memory_preservation = 6 |
|
|
|
|
|
yield None, None, '', '', gr.update(interactive=False), gr.update(interactive=True) |
|
|
|
try: |
|
stream = AsyncStream() |
|
|
|
|
|
async_run(worker, input_image, prompt, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache) |
|
|
|
output_filename = None |
|
prev_output_filename = None |
|
error_message = None |
|
|
|
|
|
while True: |
|
try: |
|
flag, data = stream.output_queue.next() |
|
|
|
if flag == 'file': |
|
output_filename = data |
|
prev_output_filename = output_filename |
|
|
|
yield output_filename, gr.update(), gr.update(), '', gr.update(interactive=False), gr.update(interactive=True) |
|
|
|
if flag == 'progress': |
|
preview, desc, html = data |
|
|
|
yield gr.update(), gr.update(visible=True, value=preview), desc, html, gr.update(interactive=False), gr.update(interactive=True) |
|
|
|
if flag == 'error': |
|
error_message = data |
|
print(f"エラーメッセージを受信: {error_message}") |
|
|
|
|
|
if flag == 'end': |
|
|
|
if output_filename is None and prev_output_filename is not None: |
|
output_filename = prev_output_filename |
|
|
|
|
|
if error_message: |
|
yield output_filename, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
else: |
|
|
|
yield output_filename, gr.update(visible=False), gr.update(), '', gr.update(interactive=True), gr.update(interactive=False) |
|
break |
|
except Exception as e: |
|
print(f"出力処理中にエラーが発生しました: {e}") |
|
|
|
current_time = time.time() |
|
if current_time - last_update_time > 60: |
|
print(f"処理がフリーズした可能性があります。{current_time - last_update_time:.1f}秒間更新がありません") |
|
|
|
|
|
if prev_output_filename: |
|
yield prev_output_filename, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
else: |
|
yield None, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
break |
|
|
|
except Exception as e: |
|
print(f"処理の開始中にエラーが発生しました: {e}") |
|
traceback.print_exc() |
|
error_msg = str(e) |
|
|
|
yield None, gr.update(visible=False), gr.update(), gr.update(interactive=True), gr.update(interactive=False) |
|
|
|
|
|
def end_process(): |
|
"""生成プロセスを停止する関数 - キューに'end'信号を送信して生成を中断します""" |
|
print("【デバッグ】ユーザーが停止ボタンをクリックしました。停止信号を送信中...") |
|
|
|
if 'stream' in globals() and stream is not None: |
|
|
|
try: |
|
current_top = stream.input_queue.top() |
|
print(f"【デバッグ】現在のキュー先頭信号: {current_top}") |
|
except Exception as e: |
|
print(f"【デバッグ】キュー状態確認中にエラー: {e}") |
|
|
|
|
|
try: |
|
stream.input_queue.push('end') |
|
print("【デバッグ】キューにend信号を正常に送信しました") |
|
|
|
|
|
try: |
|
current_top_after = stream.input_queue.top() |
|
print(f"【デバッグ】送信後のキュー先頭信号: {current_top_after}") |
|
except Exception as e: |
|
print(f"【デバッグ】送信後のキュー状態確認中にエラー: {e}") |
|
|
|
except Exception as e: |
|
print(f"【デバッグ】キューへのend信号送信に失敗: {e}") |
|
else: |
|
print("【デバッグ】警告: streamが初期化されていないため、停止信号を送信できません") |
|
return None |
|
|
|
|
|
quick_prompts = [ |
|
'The camera smoothly orbits around the center of the scene, keeping the center point fixed and always in view', |
|
] |
|
quick_prompts = [[x] for x in quick_prompts] |
|
|
|
|
|
|
|
def make_custom_css(): |
|
progress_bar_css = make_progress_bar_css() |
|
|
|
responsive_css = """ |
|
/* 基本レスポンシブ設定 */ |
|
#app-container { |
|
max-width: 100%; |
|
margin: 0 auto; |
|
} |
|
|
|
|
|
/* ページタイトルのスタイル */ |
|
h1 { |
|
font-size: 2rem; |
|
text-align: center; |
|
margin-bottom: 1rem; |
|
} |
|
|
|
/* ボタンのスタイル */ |
|
.start-btn, .stop-btn { |
|
min-height: 45px; |
|
font-size: 1rem; |
|
} |
|
|
|
/* モバイルデバイスのスタイル - 小画面 */ |
|
@media (max-width: 768px) { |
|
h1 { |
|
font-size: 1.5rem; |
|
margin-bottom: 0.5rem; |
|
} |
|
|
|
/* 単一カラムレイアウト */ |
|
.mobile-full-width { |
|
flex-direction: column !important; |
|
} |
|
|
|
.mobile-full-width > .gr-block { |
|
min-width: 100% !important; |
|
flex-grow: 1; |
|
} |
|
|
|
/* 動画サイズの調整 */ |
|
.video-container { |
|
height: auto !important; |
|
} |
|
|
|
/* ボタンサイズの調整 */ |
|
.button-container button { |
|
min-height: 50px; |
|
font-size: 1rem; |
|
touch-action: manipulation; |
|
} |
|
|
|
/* スライダーの調整 */ |
|
.slider-container input[type="range"] { |
|
height: 30px; |
|
} |
|
} |
|
|
|
/* タブレットデバイスのスタイル */ |
|
@media (min-width: 769px) and (max-width: 1024px) { |
|
.tablet-adjust { |
|
width: 48% !important; |
|
} |
|
} |
|
|
|
/* ダークモードサポート */ |
|
@media (prefers-color-scheme: dark) { |
|
.dark-mode-text { |
|
color: #f0f0f0; |
|
} |
|
|
|
.dark-mode-bg { |
|
background-color: #2a2a2a; |
|
} |
|
} |
|
|
|
/* アクセシビリティの向上 */ |
|
button, input, select, textarea { |
|
font-size: 16px; /* iOSでの拡大を防止 */ |
|
} |
|
|
|
/* タッチ操作の最適化 */ |
|
button, .interactive-element { |
|
min-height: 44px; |
|
min-width: 44px; |
|
} |
|
|
|
/* コントラストの向上 */ |
|
.high-contrast { |
|
color: #fff; |
|
background-color: #000; |
|
} |
|
|
|
/* プログレスバーのスタイル強化 */ |
|
.progress-container { |
|
margin-top: 10px; |
|
margin-bottom: 10px; |
|
} |
|
|
|
/* エラーメッセージのスタイル */ |
|
#error-message { |
|
color: #ff4444; |
|
font-weight: bold; |
|
padding: 10px; |
|
border-radius: 4px; |
|
margin-top: 10px; |
|
} |
|
|
|
/* エラーコンテナの正しい表示 */ |
|
.error-message { |
|
background-color: rgba(255, 0, 0, 0.1); |
|
padding: 10px; |
|
border-radius: 4px; |
|
margin-top: 10px; |
|
border: 1px solid #ffcccc; |
|
} |
|
|
|
/* 多言語エラーメッセージの処理 */ |
|
.error-msg-en, .error-msg-ja { |
|
font-weight: bold; |
|
} |
|
|
|
/* エラーアイコン */ |
|
.error-icon { |
|
color: #ff4444; |
|
font-size: 18px; |
|
margin-right: 8px; |
|
} |
|
|
|
/* 空のエラーメッセージが背景とボーダーを表示しないことを確認 */ |
|
#error-message:empty { |
|
background-color: transparent; |
|
border: none; |
|
padding: 0; |
|
margin: 0; |
|
} |
|
|
|
/* Gradioのデフォルトエラー表示の修正 */ |
|
.error { |
|
display: none !important; |
|
} |
|
""" |
|
|
|
|
|
combined_css = progress_bar_css + responsive_css |
|
return combined_css |
|
|
|
|
|
css = make_custom_css() |
|
block = gr.Blocks(css=css).queue() |
|
with block: |
|
gr.HTML("<h1 ='title'>FramePack_rotate_landscape - 風景画像回転動画生成</h1>") |
|
|
|
|
|
with gr.Row(elem_classes="mobile-full-width"): |
|
with gr.Column(scale=1, elem_classes="mobile-full-width"): |
|
|
|
input_image = gr.Image( |
|
sources='upload', |
|
type="numpy", |
|
label="画像をアップロード / Upload Image", |
|
elem_id="input-image", |
|
height=320 |
|
) |
|
|
|
prompt = gr.Textbox( |
|
label="プロンプト / Prompt", |
|
value='The camera smoothly orbits around the center of the scene, keeping the center point fixed and always in view', |
|
elem_id="prompt-input" |
|
) |
|
|
|
example_quick_prompts = gr.Dataset( |
|
samples=quick_prompts, |
|
label='クイックプロンプト一覧 / Quick Prompts', |
|
samples_per_page=1000, |
|
components=[prompt] |
|
) |
|
example_quick_prompts.click(lambda x: x[0], inputs=[example_quick_prompts], outputs=prompt, show_progress=False, queue=False) |
|
|
|
|
|
with gr.Row(elem_classes="button-container"): |
|
start_button = gr.Button( |
|
value="生成開始 / Generate", |
|
elem_classes="start-btn", |
|
elem_id="start-button", |
|
variant="primary" |
|
) |
|
|
|
end_button = gr.Button( |
|
value="停止 / Stop", |
|
elem_classes="stop-btn", |
|
elem_id="stop-button", |
|
interactive=False |
|
) |
|
|
|
|
|
with gr.Group(): |
|
use_teacache = gr.Checkbox( |
|
label='TeaCacheを使用 / Use TeaCache', |
|
value=True, |
|
info='処理速度が速くなりますが、指や手の生成品質が若干低下する可能性があります。 / Faster speed, but may result in slightly worse finger and hand generation.' |
|
) |
|
|
|
n_prompt = gr.Textbox(label="ネガティブプロンプト / Negative Prompt", value="", visible=False) |
|
|
|
seed = gr.Number( |
|
label="シード値 / Seed", |
|
value=31337, |
|
precision=0 |
|
) |
|
|
|
|
|
with gr.Group(elem_classes="slider-container"): |
|
total_second_length = gr.Slider( |
|
label="動画の長さ(最大3秒) / Video Length (max 3 seconds)", |
|
minimum=0.5, |
|
maximum=3, |
|
value=1, |
|
step=0.1 |
|
) |
|
|
|
|
|
with gr.Column(scale=1, elem_classes="mobile-full-width"): |
|
|
|
preview_image = gr.Image( |
|
label="プレビュー / Preview", |
|
height=200, |
|
visible=False, |
|
elem_classes="preview-container" |
|
) |
|
|
|
|
|
result_video = gr.Video( |
|
label="生成された動画 / Generated Video", |
|
autoplay=True, |
|
show_share_button=True, |
|
height=512, |
|
loop=True, |
|
elem_classes="video-container", |
|
elem_id="result-video" |
|
) |
|
|
|
gr.HTML("<div ='sampling_note' class='note'>注意:逆順サンプリングのため、終了動作が開始動作より先に生成されます。開始動作が動画に表示されていない場合は、しばらくお待ちください。後で生成されます。</div>") |
|
gr.HTML("<div ='sampling_note' class='note'>Note that the ending actions will be generated before the starting actions due to the inverted sampling. If the starting action is not in the video, you just need to wait, and it will be generated later.</div>") |
|
|
|
|
|
with gr.Group(elem_classes="progress-container"): |
|
progress_desc = gr.Markdown('', elem_classes='no-generating-animation') |
|
progress_bar = gr.HTML('', elem_classes='no-generating-animation') |
|
|
|
|
|
error_message = gr.HTML('', elem_id='error-message', visible=True) |
|
|
|
|
|
ips = [input_image, prompt, n_prompt, seed, total_second_length, use_teacache] |
|
|
|
|
|
start_button.click(fn=process, inputs=ips, outputs=[result_video, preview_image, progress_desc, progress_bar, start_button, end_button]) |
|
end_button.click(fn=end_process) |
|
|
|
|
|
block.launch() |
|
|
|
|