| 1 | import os |
| 2 | import time |
| 3 | from pathlib import Path |
| 4 | from typing import Any |
| 5 | |
| 6 | import streamlit as st |
| 7 | from loguru import logger |
| 8 | import httpx |
| 9 | from web.i18n import tr, get_language |
| 10 | from web.pipelines.base import PipelineUI, register_pipeline_ui |
| 11 | from web.pipelines.api_workflows import ( |
| 12 | list_api_media_workflows, |
| 13 | render_api_video_controls, |
| 14 | workflow_select_help, |
| 15 | workflow_source_help, |
| 16 | workflow_source_label, |
| 17 | ) |
| 18 | from web.components.content_input import render_version_info |
| 19 | from web.components.digital_tts_config import render_style_config |
| 20 | from web.utils.async_helpers import run_async |
| 21 | from web.utils.history_persistence import save_web_generation_history |
| 22 | from web.utils.streamlit_helpers import check_and_warn_selfhost_workflow |
| 23 | from pixelle_video.config import config_manager |
| 24 | from pixelle_video.utils.os_util import create_task_output_dir |
| 25 | |
| 26 | class DigitalHumanPipelineUI(PipelineUI): |
| 27 | """ |
| 28 | UI for the Digital_Human Video Generation Pipeline. |
| 29 | Generates videos from user-provided assets (images&videos&audio). |
| 30 | """ |
| 31 | name = "digital_human" |
| 32 | icon = "🤖" |
| 33 | |
| 34 | @property |
| 35 | def display_name(self): |
| 36 | return tr("pipeline.digital_human.name") |
| 37 | |
| 38 | @property |
| 39 | def description(self): |
| 40 | return tr("pipeline.digital_human.description") |
| 41 | |
| 42 | def render(self, pixelle_video: Any): |
| 43 | # Three-column layout |
| 44 | left_col, middle_col, right_col = st.columns([1, 1, 1]) |
| 45 | |
| 46 | # ==================================================================== |
| 47 | # Left Column: Asset Upload |
| 48 | # ==================================================================== |
| 49 | with left_col: |
| 50 | asset_params = self.render_digital_human_input() |
| 51 | style_params = render_style_config(pixelle_video) |
| 52 | # bgm_params = render_bgm_section(key_prefix="asset_") |
| 53 | render_version_info() |
| 54 | |
| 55 | # ==================================================================== |
| 56 | # Middle Column: Video Configuration |
| 57 | # ==================================================================== |
| 58 | with middle_col: |
| 59 | # Style configuration () |
| 60 | workflow_path = self.workflow_path_config(pixelle_video) |
| 61 | mode_params = self.render_digital_human_mode(asset_params["character_assets"]) |
| 62 | |
| 63 | # ==================================================================== |
| 64 | # Right Column: Output Preview |
| 65 | # ==================================================================== |
| 66 | with right_col: |
| 67 | # Combine all parameters |
| 68 | video_params = { |
| 69 | **mode_params, |
| 70 | **asset_params, |
| 71 | **style_params, |
| 72 | "workflow_path": workflow_path |
| 73 | } |
| 74 | |
| 75 | self._render_output_preview(pixelle_video, video_params) |
| 76 | |
| 77 | def render_digital_human_input(self) -> dict: |
| 78 | """Render digital human character image upload section""" |
| 79 | with st.container(border=True): |
| 80 | st.markdown(f"**{tr('digital_human.section.character_assets')}**") |
| 81 | |
| 82 | with st.expander(tr("help.feature_description"), expanded=False): |
| 83 | st.markdown(f"**{tr('help.what')}**") |
| 84 | st.markdown(tr("digital_human.assets.character_what")) |
| 85 | st.markdown(f"**{tr('help.how')}**") |
| 86 | st.markdown(tr("digital_human.assets.how")) |
| 87 | |
| 88 | # File uploader for multiple files |
| 89 | uploaded_files = st.file_uploader( |
| 90 | tr("digital_human.assets.upload"), |
| 91 | type=["jpg", "jpeg", "png", "webp"], |
| 92 | accept_multiple_files=True, |
| 93 | help=tr("digital_human.assets.upload_help"), |
| 94 | key="character_files" |
| 95 | ) |
| 96 | |
| 97 | # Save uploaded files to temp directory with unique session ID |
| 98 | character_asset_paths = [] |
| 99 | if uploaded_files: |
| 100 | import uuid |
| 101 | session_id = str(uuid.uuid4()).replace('-', '')[:12] |
| 102 | temp_dir = Path(f"temp/assets_{session_id}") |
| 103 | temp_dir.mkdir(parents=True, exist_ok=True) |
| 104 | |
| 105 | for uploaded_file in uploaded_files: |
| 106 | file_path = temp_dir / uploaded_file.name |
| 107 | with open(file_path, "wb") as f: |
| 108 | f.write(uploaded_file.getbuffer()) |
| 109 | character_asset_paths.append(str(file_path.absolute())) |
| 110 | |
| 111 | st.success(tr("digital_human.assets.character_sucess")) |
| 112 | |
| 113 | # Preview uploaded assets |
| 114 | with st.expander(tr("digital_human.assets.preview"), expanded=True): |
| 115 | # Show in a grid (3 columns) |
| 116 | cols = st.columns(3) |
| 117 | for i, (file, path) in enumerate(zip(uploaded_files, character_asset_paths)): |
| 118 | with cols[i % 3]: |
| 119 | # Check if image |
| 120 | ext = Path(path).suffix.lower() |
| 121 | if ext in [".jpg", ".jpeg", ".png", ".webp"]: |
| 122 | st.image(file, caption=file.name, use_container_width=True) |
| 123 | else: |
| 124 | st.info(tr("digital_human.assets.character_empty_hint")) |
| 125 | |
| 126 | return {"character_assets": character_asset_paths} |
| 127 | |
| 128 | def workflow_path_config(self, pixelle_video: Any) -> dict: |
| 129 | # Workflow source selection |
| 130 | with st.container(border=True): |
| 131 | st.markdown(f"**{tr('asset_based.section.source')}**") |
| 132 | |
| 133 | with st.expander(tr("help.feature_description"), expanded=False): |
| 134 | st.markdown(f"**{tr('help.what')}**") |
| 135 | st.markdown(tr("asset_based.source.what")) |
| 136 | st.markdown(f"**{tr('help.how')}**") |
| 137 | st.markdown(tr("asset_based.source.how")) |
| 138 | |
| 139 | source_options = { |
| 140 | "runninghub": tr("asset_based.source.runninghub"), |
| 141 | "selfhost": tr("asset_based.source.selfhost"), |
| 142 | "api": workflow_source_label("api"), |
| 143 | } |
| 144 | |
| 145 | # Check if RunningHub API key is configured |
| 146 | comfyui_config = config_manager.get_comfyui_config() |
| 147 | has_runninghub = bool(comfyui_config.get("runninghub_api_key")) |
| 148 | has_selfhost = bool(comfyui_config.get("comfyui_url")) |
| 149 | |
| 150 | workflow_config = { |
| 151 | "first_workflow_path": "workflows/runninghub/digital_image.json", |
| 152 | "second_workflow_path": "workflows/runninghub/digital_combination.json", |
| 153 | "third_workflow_path": "workflows/runninghub/digital_customize.json", |
| 154 | } |
| 155 | |
| 156 | def digital_image_workflows(source_name: str) -> list[dict]: |
| 157 | first_path = Path("workflows") / source_name / "digital_image.json" |
| 158 | third_path = Path("workflows") / source_name / "digital_customize.json" |
| 159 | if not first_path.exists() or not third_path.exists(): |
| 160 | return [] |
| 161 | return [ |
| 162 | { |
| 163 | "key": f"{source_name}/digital_image.json", |
| 164 | "display_name": f"digital_image.json - {source_name.title()}", |
| 165 | "first_workflow_path": str(first_path), |
| 166 | "third_workflow_path": str(third_path), |
| 167 | } |
| 168 | ] |
| 169 | |
| 170 | def digital_video_workflows(source_name: str) -> list[dict]: |
| 171 | second_path = Path("workflows") / source_name / "digital_combination.json" |
| 172 | if not second_path.exists(): |
| 173 | return [] |
| 174 | return [ |
| 175 | { |
| 176 | "key": f"{source_name}/digital_combination.json", |
| 177 | "display_name": f"digital_combination.json - {source_name.title()}", |
| 178 | "second_workflow_path": str(second_path), |
| 179 | } |
| 180 | ] |
| 181 | |
| 182 | api_image_workflows = list_api_media_workflows(pixelle_video, "image") |
| 183 | image_source_options = [] |
| 184 | if digital_image_workflows("runninghub"): |
| 185 | image_source_options.append("runninghub") |
| 186 | if digital_image_workflows("selfhost"): |
| 187 | image_source_options.append("selfhost") |
| 188 | if api_image_workflows: |
| 189 | image_source_options.append("api") |
| 190 | |
| 191 | workflow_config["api_image_workflow"] = None |
| 192 | if st.session_state.get("digital_human_image_service_source") not in image_source_options: |
| 193 | st.session_state.pop("digital_human_image_service_source", None) |
| 194 | image_service_source = st.radio( |
| 195 | "前置图片生成服务" if get_language() == "zh_CN" else "Pre-image generation service", |
| 196 | image_source_options, |
| 197 | format_func=lambda x: source_options[x], |
| 198 | horizontal=True, |
| 199 | key="digital_human_image_service_source", |
| 200 | help=workflow_source_help("前置图片生成" if get_language() == "zh_CN" else "pre-image generation"), |
| 201 | ) |
| 202 | |
| 203 | image_workflows = [] |
| 204 | if image_service_source in {"runninghub", "selfhost"}: |
| 205 | if image_service_source == "runninghub" and not has_runninghub: |
| 206 | st.warning(tr("asset_based.source.runninghub_not_configured")) |
| 207 | if image_service_source == "selfhost" and not has_selfhost: |
| 208 | st.warning(tr("asset_based.source.selfhost_not_configured")) |
| 209 | |
| 210 | image_workflows = digital_image_workflows(image_service_source) |
| 211 | elif image_service_source == "api": |
| 212 | if not api_image_workflows: |
| 213 | st.warning( |
| 214 | "没有找到 API 图片模型，请先配置图像模型提供商。" |
| 215 | if get_language() == "zh_CN" |
| 216 | else "No API image model found. Configure an image provider first." |
| 217 | ) |
| 218 | else: |
| 219 | image_workflows = api_image_workflows |
| 220 | |
| 221 | image_options = [wf["display_name"] for wf in image_workflows] |
| 222 | selected_image_workflow = st.selectbox( |
| 223 | "前置图片工作流/模型" if get_language() == "zh_CN" else "Pre-image workflow/model", |
| 224 | image_options if image_options else ["No workflow/model available"], |
| 225 | index=0, |
| 226 | key="digital_human_image_workflow", |
| 227 | disabled=not image_options, |
| 228 | help=workflow_select_help(), |
| 229 | ) |
| 230 | if image_options: |
| 231 | selected_index = image_options.index(selected_image_workflow) |
| 232 | selected_workflow = image_workflows[selected_index] |
| 233 | if image_service_source == "api": |
| 234 | workflow_config["api_image_workflow"] = selected_workflow["key"] |
| 235 | else: |
| 236 | workflow_config["first_workflow_path"] = selected_workflow["first_workflow_path"] |
| 237 | workflow_config["third_workflow_path"] = selected_workflow["third_workflow_path"] |
| 238 | |
| 239 | workflow_config["api_video_workflow"] = None |
| 240 | workflow_config["api_video_params"] = {} |
| 241 | api_video_workflows = list_api_media_workflows( |
| 242 | pixelle_video, |
| 243 | "video", |
| 244 | required_adapter_abilities=["digital_human"], |
| 245 | verified_only=True, |
| 246 | ) |
| 247 | video_source_options = [] |
| 248 | if digital_video_workflows("runninghub"): |
| 249 | video_source_options.append("runninghub") |
| 250 | if digital_video_workflows("selfhost"): |
| 251 | video_source_options.append("selfhost") |
| 252 | if api_video_workflows: |
| 253 | video_source_options.append("api") |
| 254 | |
| 255 | if st.session_state.get("digital_human_video_service_source") not in video_source_options: |
| 256 | st.session_state.pop("digital_human_video_service_source", None) |
| 257 | video_service_source = st.radio( |
| 258 | "口播视频合成服务" if get_language() == "zh_CN" else "Talking-video synthesis service", |
| 259 | video_source_options, |
| 260 | format_func=lambda x: source_options[x], |
| 261 | horizontal=True, |
| 262 | key="digital_human_video_service_source", |
| 263 | help=workflow_source_help("口播视频合成" if get_language() == "zh_CN" else "talking-video synthesis"), |
| 264 | ) |
| 265 | |
| 266 | video_workflows = [] |
| 267 | if video_service_source in {"runninghub", "selfhost"}: |
| 268 | if video_service_source == "runninghub" and not has_runninghub: |
| 269 | st.warning(tr("asset_based.source.runninghub_not_configured")) |
| 270 | if video_service_source == "selfhost" and not has_selfhost: |
| 271 | st.warning(tr("asset_based.source.selfhost_not_configured")) |
| 272 | |
| 273 | video_workflows = digital_video_workflows(video_service_source) |
| 274 | elif video_service_source == "api": |
| 275 | if not api_video_workflows: |
| 276 | st.warning( |
| 277 | "没有找到已验证的 API 参考生视频模型，请先配置 DashScope 等提供商。" |
| 278 | if get_language() == "zh_CN" |
| 279 | else "No verified API reference-to-video model found. Configure a provider first." |
| 280 | ) |
| 281 | else: |
| 282 | video_workflows = api_video_workflows |
| 283 | |
| 284 | video_options = [wf["display_name"] for wf in video_workflows] |
| 285 | selected_video_workflow = st.selectbox( |
| 286 | "口播视频工作流/模型" if get_language() == "zh_CN" else "Talking-video workflow/model", |
| 287 | video_options if video_options else ["No workflow/model available"], |
| 288 | index=0, |
| 289 | key="digital_human_video_workflow", |
| 290 | disabled=not video_options, |
| 291 | help=workflow_select_help(), |
| 292 | ) |
| 293 | if video_options: |
| 294 | selected_index = video_options.index(selected_video_workflow) |
| 295 | selected_workflow = video_workflows[selected_index] |
| 296 | if video_service_source == "api": |
| 297 | workflow_config["api_video_workflow"] = selected_workflow["key"] |
| 298 | workflow_config["api_video_params"] = render_api_video_controls( |
| 299 | selected_workflow, |
| 300 | key_prefix="digital_human", |
| 301 | default_duration=5, |
| 302 | ) |
| 303 | else: |
| 304 | workflow_config["second_workflow_path"] = selected_workflow["second_workflow_path"] |
| 305 | |
| 306 | missing_workflows = [ |
| 307 | path for key, path in workflow_config.items() |
| 308 | if key.endswith("_workflow_path") and isinstance(path, str) and not Path(path).exists() |
| 309 | ] |
| 310 | if missing_workflows: |
| 311 | st.warning( |
| 312 | ( |
| 313 | "当前选择缺少数字人口播工作流文件：" |
| 314 | + "、".join(missing_workflows) |
| 315 | ) |
| 316 | if get_language() == "zh_CN" |
| 317 | else "The current selection is missing digital-human workflow files: " |
| 318 | + ", ".join(missing_workflows) |
| 319 | ) |
| 320 | |
| 321 | return workflow_config |
| 322 | |
| 323 | def render_digital_human_mode(self, character_asset_paths: list) -> dict: |
| 324 | with st.container(border=True): |
| 325 | st.markdown(f"**{tr('digital_human.section.select_mode')}**") |
| 326 | |
| 327 | with st.expander(tr("help.feature_description"), expanded=False): |
| 328 | st.markdown(f"**{tr('help.what')}**") |
| 329 | st.markdown(tr("digital_human.assets.mode_what")) |
| 330 | st.markdown(f"**{tr('help.how')}**") |
| 331 | st.markdown(tr("digital_human.assets.select_how")) |
| 332 | |
| 333 | mode = st.radio( |
| 334 | "Processing Mode", |
| 335 | ["digital", "customize"], |
| 336 | horizontal=True, |
| 337 | format_func=lambda x: tr(f"mode.{x}"), |
| 338 | label_visibility="collapsed", |
| 339 | key="mode_selection" |
| 340 | ) |
| 341 | |
| 342 | # Text input (unified for both modes) |
| 343 | text_placeholder = tr("digital_human.input.topic_placeholder") if mode == "digital" else tr("digital_human.input.content_placeholder") |
| 344 | text_height = 120 if mode == "digital" else 200 |
| 345 | text_help = tr("input.text_help_digital") if mode == "digital" else tr("input.text_help_fixed") |
| 346 | |
| 347 | if mode == "digital": |
| 348 | # File uploader for multiple files |
| 349 | uploaded_files = st.file_uploader( |
| 350 | tr("digital_human.assets.upload"), |
| 351 | type=["jpg", "jpeg", "png", "webp"], |
| 352 | accept_multiple_files=True, |
| 353 | help=tr("digital_human.assets.upload_help"), |
| 354 | key="digital_files" |
| 355 | ) |
| 356 | |
| 357 | # Save uploaded files to temp directory with unique session ID |
| 358 | goods_asset_paths = [] |
| 359 | if uploaded_files: |
| 360 | import uuid |
| 361 | session_id = str(uuid.uuid4()).replace('-', '')[:12] |
| 362 | temp_dir = Path(f"temp/assets_{session_id}") |
| 363 | temp_dir.mkdir(parents=True, exist_ok=True) |
| 364 | |
| 365 | for uploaded_file in uploaded_files: |
| 366 | file_path = temp_dir / uploaded_file.name |
| 367 | with open(file_path, "wb") as f: |
| 368 | f.write(uploaded_file.getbuffer()) |
| 369 | goods_asset_paths.append(str(file_path.absolute())) |
| 370 | |
| 371 | st.success(tr("digital_human.assets.goods_sucess")) |
| 372 | |
| 373 | # Preview uploaded assets |
| 374 | with st.expander(tr("digital_human.assets.preview"), expanded=True): |
| 375 | # Show in a grid (3 columns) |
| 376 | cols = st.columns(3) |
| 377 | for i, (file, path) in enumerate(zip(uploaded_files, goods_asset_paths)): |
| 378 | with cols[i % 3]: |
| 379 | # Check if image |
| 380 | ext = Path(path).suffix.lower() |
| 381 | if ext in [".jpg", ".jpeg", ".png", ".webp"]: |
| 382 | st.image(file, caption=file.name, use_container_width=True) |
| 383 | else: |
| 384 | st.info(tr("digital_human.assets.goods_empty_hint")) |
| 385 | # Text input |
| 386 | goods_text = st.text_area( |
| 387 | tr("digital_human.input_text"), |
| 388 | placeholder=text_placeholder, |
| 389 | height=text_height, |
| 390 | help=text_help, |
| 391 | key="digital_box" |
| 392 | ) |
| 393 | |
| 394 | goods_title = st.text_input( |
| 395 | tr("digital_human.goods_title"), |
| 396 | placeholder=tr("digital_human.goods_title_placeholder"), |
| 397 | help=tr("digital_human.goods_title_help"), |
| 398 | key="goods_title" |
| 399 | ) |
| 400 | |
| 401 | return { |
| 402 | "character_assets": character_asset_paths, |
| 403 | "goods_title": goods_title, |
| 404 | "goods_assets": goods_asset_paths, |
| 405 | "goods_text": goods_text, |
| 406 | "mode": mode |
| 407 | } |
| 408 | |
| 409 | else: |
| 410 | goods_text = st.text_area( |
| 411 | tr("digital_human.customize_text"), |
| 412 | placeholder=text_placeholder, |
| 413 | height=text_height, |
| 414 | help=text_help, |
| 415 | key="customize_box" |
| 416 | ) |
| 417 | |
| 418 | return { |
| 419 | "character_assets": character_asset_paths, |
| 420 | "goods_text": goods_text, |
| 421 | "mode": mode |
| 422 | } |
| 423 | |
| 424 | def _render_output_preview(self, pixelle_video: Any, video_params: dict): |
| 425 | """Render output preview section""" |
| 426 | with st.container(border=True): |
| 427 | st.markdown(f"**{tr('section.video_generation')}**") |
| 428 | |
| 429 | # Check configuration |
| 430 | if not config_manager.validate(): |
| 431 | st.warning(tr("settings.not_configured")) |
| 432 | |
| 433 | # Get input data |
| 434 | character_assets = video_params.get("character_assets", []) |
| 435 | goods_assets = video_params.get("goods_assets", []) |
| 436 | goods_title = video_params.get("goods_title", "") |
| 437 | goods_text = video_params.get("goods_text", "") |
| 438 | mode = video_params.get("mode") |
| 439 | tts_voice = video_params.get("tts_voice", "zh-CN-YunjianNeural") |
| 440 | tts_speed = video_params.get("tts_speed", 1.2) |
| 441 | |
| 442 | logger.info(f"🔧 The obtained TTS parameters:") |
| 443 | logger.info(f" - tts_voice: {tts_voice}") |
| 444 | logger.info(f" - tts_speed: {tts_speed}") |
| 445 | logger.info(f" - video_params中的tts_voice: {video_params.get('tts_voice', 'NOT_FOUND')}") |
| 446 | logger.info(f" - video_params: {video_params}") |
| 447 | |
| 448 | # Validation |
| 449 | if not character_assets: |
| 450 | st.info(tr("digital_human.assets.character_warning")) |
| 451 | st.button( |
| 452 | tr("btn.generate"), |
| 453 | type="primary", |
| 454 | use_container_width=True, |
| 455 | disabled=True, |
| 456 | key="digital_human_generate_disabled" |
| 457 | ) |
| 458 | return |
| 459 | |
| 460 | if mode == "digital" and not goods_assets: |
| 461 | st.info(tr("digital_human.assets.goods_warning")) |
| 462 | st.button( |
| 463 | tr("btn.generate"), |
| 464 | type="primary", |
| 465 | use_container_width=True, |
| 466 | disabled=True, |
| 467 | key="digital_human_goods_vaiidation" |
| 468 | ) |
| 469 | return |
| 470 | |
| 471 | if mode == "digital" and not (goods_text or goods_title): |
| 472 | st.info(tr("digital_human.assets.digital_mode")) |
| 473 | st.button( |
| 474 | tr("btn.generate"), |
| 475 | type="primary", |
| 476 | use_container_width=True, |
| 477 | disabled=True, |
| 478 | key="digital_human_digital_disable" |
| 479 | ) |
| 480 | return |
| 481 | |
| 482 | if mode == "digital" and (goods_text or goods_title): |
| 483 | st.warning(tr("digital_human.assets.digital_mode_warning")) |
| 484 | |
| 485 | if mode == "customize" and not goods_text: |
| 486 | st.info(tr("digital_human.assets.customize_mode")) |
| 487 | st.button( |
| 488 | tr("btn.generate"), |
| 489 | type="primary", |
| 490 | use_container_width=True, |
| 491 | disabled=True, |
| 492 | key="digital_human_customize_disable" |
| 493 | ) |
| 494 | return |
| 495 | |
| 496 | # Generate button |
| 497 | if st.button(tr("btn.generate"), type="primary", use_container_width=True, key="digital_human_generate"): |
| 498 | # Validate |
| 499 | if not config_manager.validate(): |
| 500 | st.error(tr("settings.not_configured")) |
| 501 | st.stop() |
| 502 | |
| 503 | # Show progress |
| 504 | progress_bar = st.progress(0) |
| 505 | status_text = st.empty() |
| 506 | |
| 507 | start_time = time.time() |
| 508 | |
| 509 | try: |
| 510 | # Define async generation function |
| 511 | async def generate_digital_human_video(): |
| 512 | task_dir, task_id = create_task_output_dir() |
| 513 | workflow_path = video_params["workflow_path"] |
| 514 | api_video_workflow = workflow_path.get("api_video_workflow") |
| 515 | api_video_params = dict(workflow_path.get("api_video_params") or {}) |
| 516 | |
| 517 | import json |
| 518 | from pathlib import Path |
| 519 | |
| 520 | async def generate_tts_reference(text: str) -> str: |
| 521 | audio_path = os.path.join(task_dir, "narration.mp3") |
| 522 | tts_inference_mode = video_params.get("tts_inference_mode", "local") |
| 523 | tts_voice = video_params.get("tts_voice") |
| 524 | tts_speed = video_params.get("tts_speed") |
| 525 | tts_workflow = video_params.get("tts_workflow") |
| 526 | ref_audio = video_params.get("ref_audio") |
| 527 | |
| 528 | tts_kwargs = { |
| 529 | "text": text, |
| 530 | "output_path": audio_path, |
| 531 | "inference_mode": tts_inference_mode, |
| 532 | } |
| 533 | if tts_inference_mode == "local": |
| 534 | tts_kwargs["voice"] = tts_voice |
| 535 | tts_kwargs["speed"] = tts_speed |
| 536 | elif tts_inference_mode == "comfyui": |
| 537 | if tts_workflow: |
| 538 | tts_kwargs["workflow"] = tts_workflow |
| 539 | if ref_audio: |
| 540 | tts_kwargs["ref_audio"] = ref_audio |
| 541 | |
| 542 | await pixelle_video.tts(**tts_kwargs) |
| 543 | return audio_path |
| 544 | |
| 545 | async def generate_api_digital_human(text: str) -> str: |
| 546 | status_text.text(tr("progress.step_audio")) |
| 547 | progress_bar.progress(25) |
| 548 | audio_path = await generate_tts_reference(text) |
| 549 | |
| 550 | reference_image_paths = [character_assets[0]] |
| 551 | if mode == "digital" and goods_assets: |
| 552 | reference_image_paths.append(goods_assets[0]) |
| 553 | |
| 554 | subject_prompt = ( |
| 555 | "参考图1中的人物面对镜头自然口播。" |
| 556 | if get_language() == "zh_CN" |
| 557 | else "The person in reference image 1 speaks naturally to camera." |
| 558 | ) |
| 559 | if mode == "digital" and goods_assets: |
| 560 | subject_prompt += ( |
| 561 | "结合参考图2中的商品，生成竖屏商业口播视频。" |
| 562 | if get_language() == "zh_CN" |
| 563 | else "Use the product in reference image 2 and create a vertical product-promotion talking video." |
| 564 | ) |
| 565 | prompt = f"{subject_prompt} 口播文案：{text}" |
| 566 | |
| 567 | final_video_path = os.path.join(task_dir, "final.mp4") |
| 568 | duration = int(api_video_params.pop("duration", 5)) |
| 569 | media_params = { |
| 570 | **api_video_params, |
| 571 | "prompt": prompt, |
| 572 | "workflow": api_video_workflow, |
| 573 | "media_type": "video", |
| 574 | "output_path": final_video_path, |
| 575 | "duration": duration, |
| 576 | "reference_image_paths": reference_image_paths, |
| 577 | "reference_audio_path": audio_path, |
| 578 | "audio": True, |
| 579 | "video_ratio": api_video_params.get("video_ratio", "9:16"), |
| 580 | } |
| 581 | progress_bar.progress(60) |
| 582 | status_text.text(tr("progress.generation")) |
| 583 | media_result = await pixelle_video.media(**media_params) |
| 584 | progress_bar.progress(100) |
| 585 | status_text.text(tr("status.success")) |
| 586 | return media_result.url |
| 587 | |
| 588 | if api_video_workflow: |
| 589 | if mode == "customize": |
| 590 | generated_text = goods_text |
| 591 | elif goods_text and goods_text.strip(): |
| 592 | generated_text = goods_text |
| 593 | else: |
| 594 | generated_text = await pixelle_video.llm( |
| 595 | prompt=( |
| 596 | f"请为商品“{goods_title}”写一段适合数字人口播短视频的中文推广文案。" |
| 597 | "要求自然、有吸引力，控制在80字以内，只输出文案正文。" |
| 598 | ), |
| 599 | temperature=0.7, |
| 600 | max_tokens=300, |
| 601 | ) |
| 602 | return await generate_api_digital_human(generated_text) |
| 603 | |
| 604 | kit = await pixelle_video._get_or_create_comfykit() |
| 605 | |
| 606 | if mode == "customize": |
| 607 | status_text.text(tr("progress.step_audio")) |
| 608 | progress_bar.progress(25) |
| 609 | generated_image_path = character_assets[0] |
| 610 | generated_text = goods_text |
| 611 | |
| 612 | # TTS |
| 613 | audio_path = os.path.join(task_dir, "narration.mp3") |
| 614 | tts_inference_mode = video_params.get("tts_inference_mode", "local") |
| 615 | tts_voice = video_params.get("tts_voice") |
| 616 | tts_speed = video_params.get("tts_speed") |
| 617 | tts_workflow = video_params.get("tts_workflow") |
| 618 | ref_audio = video_params.get("ref_audio") |
| 619 | |
| 620 | tts_kwargs = { |
| 621 | "text": generated_text, |
| 622 | "output_path": audio_path, |
| 623 | "inference_mode": tts_inference_mode |
| 624 | } |
| 625 | if tts_inference_mode == "local": |
| 626 | tts_kwargs["voice"] = tts_voice |
| 627 | tts_kwargs["speed"] = tts_speed |
| 628 | elif tts_inference_mode == "comfyui": |
| 629 | if tts_workflow: |
| 630 | tts_kwargs["workflow"] = tts_workflow |
| 631 | if ref_audio: |
| 632 | tts_kwargs["ref_audio"] = ref_audio |
| 633 | |
| 634 | await pixelle_video.tts(**tts_kwargs) |
| 635 | progress_bar.progress(65) |
| 636 | status_text.text(tr("progress.concatenating")) |
| 637 | |
| 638 | # Directly call the second workflow |
| 639 | second_workflow_path = Path(workflow_path.get("second_workflow_path")) |
| 640 | if not second_workflow_path.exists(): |
| 641 | raise Exception(f"The second step workflow file does not exist:{second_workflow_path}") |
| 642 | with open(second_workflow_path, 'r', encoding='utf-8') as f: |
| 643 | second_workflow_config = json.load(f) |
| 644 | second_workflow_params = { |
| 645 | "videoimage": generated_image_path, |
| 646 | "audio": audio_path |
| 647 | } |
| 648 | if second_workflow_config.get("source") == "runninghub" and "workflow_id" in second_workflow_config: |
| 649 | workflow_input = second_workflow_config["workflow_id"] |
| 650 | else: |
| 651 | workflow_input = str(second_workflow_config) |
| 652 | second_result = await kit.execute(workflow_input, second_workflow_params) |
| 653 | # Video Link Extraction |
| 654 | generated_video_url = None |
| 655 | if hasattr(second_result, 'videos') and second_result.videos: |
| 656 | generated_video_url = second_result.videos[0] |
| 657 | elif hasattr(second_result, 'outputs') and second_result.outputs: |
| 658 | for node_id, node_output in second_result.outputs.items(): |
| 659 | if isinstance(node_output, dict) and 'videos' in node_output: |
| 660 | videos = node_output['videos'] |
| 661 | if videos and len(videos) > 0: |
| 662 | generated_video_url = videos[0] |
| 663 | break |
| 664 | if not generated_video_url: |
| 665 | raise Exception("The second step of the workflow did not return a video. Please check the workflow configuration.") |
| 666 | |
| 667 | final_video_path = os.path.join(task_dir, "final.mp4") |
| 668 | timeout = httpx.Timeout(300.0) |
| 669 | async with httpx.AsyncClient(timeout=timeout) as client: |
| 670 | response = await client.get(generated_video_url) |
| 671 | response.raise_for_status() |
| 672 | with open(final_video_path, 'wb') as f: |
| 673 | f.write(response.content) |
| 674 | progress_bar.progress(100) |
| 675 | status_text.text(tr("status.success")) |
| 676 | return final_video_path |
| 677 | |
| 678 | else: |
| 679 | #Initialization and parameter preparation |
| 680 | task_dir, task_id = create_task_output_dir() |
| 681 | logger.info(f"[Initialization] Task Directory: {task_dir}") |
| 682 | |
| 683 | first_workflow_path = Path(workflow_path.get("first_workflow_path")) |
| 684 | third_workflow_path = Path(workflow_path.get("third_workflow_path")) |
| 685 | second_workflow_path = Path(workflow_path.get("second_workflow_path")) |
| 686 | api_image_workflow = workflow_path.get("api_image_workflow") |
| 687 | assert first_workflow_path.exists(), "The first_workflow file does not exist." |
| 688 | assert third_workflow_path.exists(), "The third_workflow file does not exist." |
| 689 | assert second_workflow_path.exists(), "The second_workflow file does not exist." |
| 690 | |
| 691 | if goods_text and goods_text.strip(): |
| 692 | generated_text = goods_text |
| 693 | |
| 694 | status_text.text(tr("progress.step_image")) |
| 695 | if api_image_workflow: |
| 696 | image_prompt = ( |
| 697 | f"Create a polished digital-human product promotion image. " |
| 698 | f"Use the first reference image as the person/character, the second reference image as the product, " |
| 699 | f"and make the scene suitable for a short spoken ad. Script: {goods_text}" |
| 700 | ) |
| 701 | generated_image_path = os.path.join(task_dir, "generated_digital_image.png") |
| 702 | media_result = await pixelle_video.media( |
| 703 | prompt=image_prompt, |
| 704 | workflow=api_image_workflow, |
| 705 | media_type="image", |
| 706 | image_paths=[character_assets[0], goods_assets[0]], |
| 707 | output_path=generated_image_path, |
| 708 | width=1080, |
| 709 | height=1920, |
| 710 | ) |
| 711 | generated_image_url = media_result.url |
| 712 | else: |
| 713 | workflow_path = third_workflow_path |
| 714 | workflow_params = {"firstimage": character_assets[0], "secondimage": goods_assets[0]} |
| 715 | kit = await pixelle_video._get_or_create_comfykit() |
| 716 | workflow_config = json.load(open(workflow_path, 'r', encoding='utf8')) |
| 717 | if workflow_config.get("source") == "runninghub" and "workflow_id" in workflow_config: |
| 718 | workflow_input = workflow_config["workflow_id"] |
| 719 | else: |
| 720 | workflow_input = str(workflow_config) |
| 721 | combine_image = await kit.execute(workflow_input, workflow_params) |
| 722 | if combine_image.status != "completed": |
| 723 | raise Exception(f"workflow execution failed: {combine_image.msg}") |
| 724 | generated_image_url = getattr(combine_image, "images", [None])[0] |
| 725 | status_text.text(tr("progress.step_audio")) |
| 726 | audio_path = os.path.join(task_dir, "narration.mp3") |
| 727 | tts_inference_mode = video_params.get("tts_inference_mode", "local") |
| 728 | tts_voice = video_params.get("tts_voice") |
| 729 | tts_speed = video_params.get("tts_speed") |
| 730 | tts_workflow = video_params.get("tts_workflow") |
| 731 | ref_audio = video_params.get("ref_audio") |
| 732 | |
| 733 | tts_kwargs = { |
| 734 | "text": generated_text, |
| 735 | "output_path": audio_path, |
| 736 | "inference_mode": tts_inference_mode |
| 737 | } |
| 738 | if tts_inference_mode == "local": |
| 739 | tts_kwargs["voice"] = tts_voice |
| 740 | tts_kwargs["speed"] = tts_speed |
| 741 | elif tts_inference_mode == "comfyui": |
| 742 | if tts_workflow: |
| 743 | tts_kwargs["workflow"] = tts_workflow |
| 744 | if ref_audio: |
| 745 | tts_kwargs["ref_audio"] = ref_audio |
| 746 | |
| 747 | await pixelle_video.tts(**tts_kwargs) |
| 748 | progress_bar.progress(65) |
| 749 | status_text.text(tr("progress.concatenating")) |
| 750 | |
| 751 | if not second_workflow_path.exists(): |
| 752 | raise Exception(f"The second step workflow file does not exist:{second_workflow_path}") |
| 753 | with open(second_workflow_path, 'r', encoding='utf-8') as f: |
| 754 | second_workflow_config = json.load(f) |
| 755 | second_workflow_params = { |
| 756 | "videoimage": generated_image_url, |
| 757 | "audio": audio_path |
| 758 | } |
| 759 | if second_workflow_config.get("source") == "runninghub" and "workflow_id" in second_workflow_config: |
| 760 | workflow_input = second_workflow_config["workflow_id"] |
| 761 | else: |
| 762 | workflow_input = str(second_workflow_config) |
| 763 | second_result = await kit.execute(workflow_input, second_workflow_params) |
| 764 | # Video Link Extraction |
| 765 | generated_video_url = None |
| 766 | if hasattr(second_result, 'videos') and second_result.videos: |
| 767 | generated_video_url = second_result.videos[0] |
| 768 | elif hasattr(second_result, 'outputs') and second_result.outputs: |
| 769 | for node_id, node_output in second_result.outputs.items(): |
| 770 | if isinstance(node_output, dict) and 'videos' in node_output: |
| 771 | videos = node_output['videos'] |
| 772 | if videos and len(videos) > 0: |
| 773 | generated_video_url = videos[0] |
| 774 | break |
| 775 | if not generated_video_url: |
| 776 | raise Exception("The second step of the workflow did not return a video. Please check the workflow configuration.") |
| 777 | |
| 778 | final_video_path = os.path.join(task_dir, "final.mp4") |
| 779 | timeout = httpx.Timeout(300.0) |
| 780 | async with httpx.AsyncClient(timeout=timeout) as client: |
| 781 | response = await client.get(generated_video_url) |
| 782 | response.raise_for_status() |
| 783 | with open(final_video_path, 'wb') as f: |
| 784 | f.write(response.content) |
| 785 | progress_bar.progress(100) |
| 786 | status_text.text(tr("status.success")) |
| 787 | return final_video_path |
| 788 | |
| 789 | else: |
| 790 | status_text.text(tr("progress.step_image")) |
| 791 | if api_image_workflow: |
| 792 | image_prompt = ( |
| 793 | f"Create a polished digital-human product promotion image for '{goods_title}'. " |
| 794 | f"Use the first reference image as the person/character and the second reference image as the product. " |
| 795 | f"Make it vertical, clean, commercial, and suitable for a spoken short video." |
| 796 | ) |
| 797 | generated_image_path = os.path.join(task_dir, "generated_digital_image.png") |
| 798 | media_result = await pixelle_video.media( |
| 799 | prompt=image_prompt, |
| 800 | workflow=api_image_workflow, |
| 801 | media_type="image", |
| 802 | image_paths=[character_assets[0], goods_assets[0]], |
| 803 | output_path=generated_image_path, |
| 804 | width=1080, |
| 805 | height=1920, |
| 806 | ) |
| 807 | generated_image_url = media_result.url |
| 808 | generated_text = await pixelle_video.llm( |
| 809 | prompt=( |
| 810 | f"请为商品“{goods_title}”写一段适合数字人口播短视频的中文推广文案。" |
| 811 | "要求自然、有吸引力，控制在80字以内，只输出文案正文。" |
| 812 | ), |
| 813 | temperature=0.7, |
| 814 | max_tokens=300, |
| 815 | ) |
| 816 | else: |
| 817 | workflow_path = first_workflow_path |
| 818 | workflow_params = {"firstimage": character_assets[0], "secondimage": goods_assets[0], "goodstype": goods_title} |
| 819 | kit = await pixelle_video._get_or_create_comfykit() |
| 820 | workflow_config = json.load(open(workflow_path, 'r', encoding='utf8')) |
| 821 | if workflow_config.get("source") == "runninghub" and "workflow_id" in workflow_config: |
| 822 | workflow_input = workflow_config["workflow_id"] |
| 823 | else: |
| 824 | workflow_input = str(workflow_config) |
| 825 | synthesis_result = await kit.execute(workflow_input, workflow_params) |
| 826 | if synthesis_result.status != "completed": |
| 827 | raise Exception(f"workflow execution failed: {synthesis_result.msg}") |
| 828 | generated_image_url = getattr(synthesis_result, "images", [None])[0] |
| 829 | generated_text = getattr(synthesis_result, "texts", [None])[0] |
| 830 | |
| 831 | status_text.text(tr("progress.step_audio")) |
| 832 | audio_path = os.path.join(task_dir, "narration.mp3") |
| 833 | tts_inference_mode = video_params.get("tts_inference_mode", "local") |
| 834 | tts_voice = video_params.get("tts_voice") |
| 835 | tts_speed = video_params.get("tts_speed") |
| 836 | tts_workflow = video_params.get("tts_workflow") |
| 837 | ref_audio = video_params.get("ref_audio") |
| 838 | |
| 839 | tts_kwargs = { |
| 840 | "text": generated_text, |
| 841 | "output_path": audio_path, |
| 842 | "inference_mode": tts_inference_mode |
| 843 | } |
| 844 | if tts_inference_mode == "local": |
| 845 | tts_kwargs["voice"] = tts_voice |
| 846 | tts_kwargs["speed"] = tts_speed |
| 847 | elif tts_inference_mode == "comfyui": |
| 848 | if tts_workflow: |
| 849 | tts_kwargs["workflow"] = tts_workflow |
| 850 | if ref_audio: |
| 851 | tts_kwargs["ref_audio"] = ref_audio |
| 852 | |
| 853 | await pixelle_video.tts(**tts_kwargs) |
| 854 | progress_bar.progress(65) |
| 855 | status_text.text(tr("progress.concatenating")) |
| 856 | |
| 857 | if not second_workflow_path.exists(): |
| 858 | raise Exception(f"The second step workflow file does not exist:{second_workflow_path}") |
| 859 | with open(second_workflow_path, 'r', encoding='utf-8') as f: |
| 860 | second_workflow_config = json.load(f) |
| 861 | second_workflow_params = { |
| 862 | "videoimage": generated_image_url, |
| 863 | "audio": audio_path |
| 864 | } |
| 865 | if second_workflow_config.get("source") == "runninghub" and "workflow_id" in second_workflow_config: |
| 866 | workflow_input = second_workflow_config["workflow_id"] |
| 867 | else: |
| 868 | workflow_input = str(second_workflow_config) |
| 869 | second_result = await kit.execute(workflow_input, second_workflow_params) |
| 870 | # Video Link Extraction |
| 871 | generated_video_url = None |
| 872 | if hasattr(second_result, 'videos') and second_result.videos: |
| 873 | generated_video_url = second_result.videos[0] |
| 874 | elif hasattr(second_result, 'outputs') and second_result.outputs: |
| 875 | for node_id, node_output in second_result.outputs.items(): |
| 876 | if isinstance(node_output, dict) and 'videos' in node_output: |
| 877 | videos = node_output['videos'] |
| 878 | if videos and len(videos) > 0: |
| 879 | generated_video_url = videos[0] |
| 880 | break |
| 881 | if not generated_video_url: |
| 882 | raise Exception("The second step of the workflow did not return a video. Please check the workflow configuration.") |
| 883 | |
| 884 | final_video_path = os.path.join(task_dir, "final.mp4") |
| 885 | timeout = httpx.Timeout(300.0) |
| 886 | async with httpx.AsyncClient(timeout=timeout) as client: |
| 887 | response = await client.get(generated_video_url) |
| 888 | response.raise_for_status() |
| 889 | with open(final_video_path, 'wb') as f: |
| 890 | f.write(response.content) |
| 891 | progress_bar.progress(100) |
| 892 | status_text.text(tr("status.success")) |
| 893 | return final_video_path |
| 894 | |
| 895 | # Execute async generation |
| 896 | final_video_path = run_async(generate_digital_human_video()) |
| 897 | run_async(save_web_generation_history( |
| 898 | pixelle_video, |
| 899 | task_id=Path(final_video_path).parent.name, |
| 900 | video_path=final_video_path, |
| 901 | pipeline="digital_human", |
| 902 | title="数字人口播" if get_language() == "zh_CN" else "Digital Human", |
| 903 | input_params={ |
| 904 | "text": goods_text or goods_title, |
| 905 | "mode": mode, |
| 906 | "goods_title": goods_title, |
| 907 | "goods_text": goods_text, |
| 908 | "character_assets": character_assets, |
| 909 | "goods_assets": goods_assets, |
| 910 | "workflow_path": video_params.get("workflow_path"), |
| 911 | "tts_voice": video_params.get("tts_voice"), |
| 912 | "tts_speed": video_params.get("tts_speed"), |
| 913 | "tts_inference_mode": video_params.get("tts_inference_mode"), |
| 914 | }, |
| 915 | )) |
| 916 | |
| 917 | total_time = time.time() - start_time |
| 918 | progress_bar.progress(100) |
| 919 | status_text.text(tr("status.success")) |
| 920 | |
| 921 | # Display result |
| 922 | st.success(tr("status.video_generated", path=final_video_path)) |
| 923 | |
| 924 | st.markdown("---") |
| 925 | |
| 926 | # Video info |
| 927 | if os.path.exists(final_video_path): |
| 928 | file_size_mb = os.path.getsize(final_video_path) / (1024 * 1024) |
| 929 | |
| 930 | info_text = ( |
| 931 | f"⏱️ {tr('info.generation_time')} {total_time:.1f}s " |
| 932 | f"📦 {file_size_mb:.2f}MB" |
| 933 | ) |
| 934 | st.caption(info_text) |
| 935 | |
| 936 | st.markdown("---") |
| 937 | |
| 938 | # Video preview |
| 939 | st.video(final_video_path) |
| 940 | |
| 941 | # Download button |
| 942 | with open(final_video_path, "rb") as video_file: |
| 943 | video_bytes = video_file.read() |
| 944 | video_filename = os.path.basename(final_video_path) |
| 945 | st.download_button( |
| 946 | label="⬇️ 下载视频" if get_language() == "zh_CN" else "⬇️ Download Video", |
| 947 | data=video_bytes, |
| 948 | file_name=video_filename, |
| 949 | mime="video/mp4", |
| 950 | use_container_width=True |
| 951 | ) |
| 952 | else: |
| 953 | st.error(tr("status.video_not_found", path=final_video_path)) |
| 954 | |
| 955 | except Exception as e: |
| 956 | status_text.text("") |
| 957 | progress_bar.empty() |
| 958 | st.error(tr("status.error", error=str(e))) |
| 959 | logger.exception(e) |
| 960 | st.stop() |
| 961 | |
| 962 | |
| 963 | # Register self |
| 964 | register_pipeline_ui(DigitalHumanPipelineUI) |
| 965 |