| 1 | import os |
| 2 | import shutil |
| 3 | import logging |
| 4 | from agents import Screenwriter, CharacterExtractor, CharacterPortraitsGenerator |
| 5 | from pipelines.script2video_pipeline import Script2VideoPipeline, _scoped_progress |
| 6 | from interfaces import CharacterInScene |
| 7 | from typing import Any, Callable, List, Dict, Optional |
| 8 | import asyncio |
| 9 | import json |
| 10 | import yaml |
| 11 | from langchain.chat_models import init_chat_model |
| 12 | from tools.render_backend import RenderBackend |
| 13 | from utils.provider_presets import resolve_chat_model_config |
| 14 | from utils.text import safe_path_component |
| 15 | from utils.video import concatenate_video_files |
| 16 | from utils.image_selection import DEFAULT_IMAGE_CANDIDATES, image_candidate_count_from_config, validate_image_candidate_count |
| 17 | |
| 18 | |
| 19 | def _pipeline_print(quiet: bool, message: str) -> None: |
| 20 | if not quiet: |
| 21 | print(message) |
| 22 | |
| 23 | |
| 24 | class Idea2VideoPipeline: |
| 25 | def __init__( |
| 26 | self, |
| 27 | chat_model: str, |
| 28 | image_generator: str, |
| 29 | video_generator: str, |
| 30 | working_dir: str, |
| 31 | num_image_candidates: int = DEFAULT_IMAGE_CANDIDATES, |
| 32 | ): |
| 33 | self.chat_model = chat_model |
| 34 | self.image_generator = image_generator |
| 35 | self.video_generator = video_generator |
| 36 | self.num_image_candidates = validate_image_candidate_count(num_image_candidates) |
| 37 | self.working_dir = working_dir |
| 38 | os.makedirs(self.working_dir, exist_ok=True) |
| 39 | |
| 40 | self.screenwriter = Screenwriter(chat_model=self.chat_model) |
| 41 | self.character_extractor = CharacterExtractor( |
| 42 | chat_model=self.chat_model) |
| 43 | self.character_portraits_generator = CharacterPortraitsGenerator( |
| 44 | image_generator=self.image_generator) |
| 45 | |
| 46 | @classmethod |
| 47 | def init_from_config(cls, config_path: str): |
| 48 | with open(config_path, "r") as f: |
| 49 | config = yaml.safe_load(f) |
| 50 | |
| 51 | chat_model_args = resolve_chat_model_config(config["chat_model"]["init_args"]) |
| 52 | chat_model = init_chat_model(**chat_model_args) |
| 53 | backend = RenderBackend.from_config(config) |
| 54 | |
| 55 | return cls( |
| 56 | chat_model=chat_model, |
| 57 | image_generator=backend.image_generator, |
| 58 | video_generator=backend.video_generator, |
| 59 | working_dir=config["working_dir"], |
| 60 | num_image_candidates=image_candidate_count_from_config(config), |
| 61 | ) |
| 62 | |
| 63 | async def extract_characters( |
| 64 | self, |
| 65 | story: str, |
| 66 | quiet: bool = False, |
| 67 | ): |
| 68 | save_path = os.path.join(self.working_dir, "characters.json") |
| 69 | |
| 70 | if os.path.exists(save_path): |
| 71 | with open(save_path, "r", encoding="utf-8") as f: |
| 72 | characters = json.load(f) |
| 73 | characters = [CharacterInScene.model_validate( |
| 74 | character) for character in characters] |
| 75 | _pipeline_print(quiet, f"🚀 Loaded {len(characters)} characters from existing file.") |
| 76 | else: |
| 77 | characters = await self.character_extractor.extract_characters(story) |
| 78 | with open(save_path, "w", encoding="utf-8") as f: |
| 79 | json.dump([character.model_dump() |
| 80 | for character in characters], f, ensure_ascii=False, indent=4) |
| 81 | _pipeline_print(quiet, f"✅ Extracted {len(characters)} characters from story and saved to {save_path}.") |
| 82 | |
| 83 | return characters |
| 84 | |
| 85 | async def generate_character_portraits( |
| 86 | self, |
| 87 | characters: List[CharacterInScene], |
| 88 | character_portraits_registry: Optional[Dict[str, Dict[str, Dict[str, str]]]], |
| 89 | style: str, |
| 90 | ): |
| 91 | character_portraits_registry_path = os.path.join( |
| 92 | self.working_dir, "character_portraits_registry.json") |
| 93 | if character_portraits_registry is None: |
| 94 | if os.path.exists(character_portraits_registry_path): |
| 95 | with open(character_portraits_registry_path, 'r', encoding='utf-8') as f: |
| 96 | character_portraits_registry = json.load(f) |
| 97 | else: |
| 98 | character_portraits_registry = {} |
| 99 | |
| 100 | tasks = [ |
| 101 | self.generate_portraits_for_single_character(character, style) |
| 102 | for character in characters |
| 103 | if character.identifier_in_scene not in character_portraits_registry |
| 104 | # Characters never shown on screen (e.g. a voice or chat-only |
| 105 | # character) have no physical description, so asking the image |
| 106 | # model for front/side/back portraits of them is nonsensical and |
| 107 | # fails repeatedly (finish_reason=IMAGE_OTHER, empty candidates). |
| 108 | and character.is_visible |
| 109 | ] |
| 110 | if tasks: |
| 111 | for future in asyncio.as_completed(tasks): |
| 112 | character_portraits_registry.update(await future) |
| 113 | with open(character_portraits_registry_path, 'w', encoding='utf-8') as f: |
| 114 | json.dump(character_portraits_registry, |
| 115 | f, ensure_ascii=False, indent=4) |
| 116 | |
| 117 | print( |
| 118 | f"✅ Completed character portrait generation for {len(characters)} characters.") |
| 119 | else: |
| 120 | print( |
| 121 | "🚀 All characters already have portraits, skipping portrait generation.") |
| 122 | |
| 123 | return character_portraits_registry |
| 124 | |
| 125 | async def develop_story( |
| 126 | self, |
| 127 | idea: str, |
| 128 | user_requirement: str, |
| 129 | quiet: bool = False, |
| 130 | ): |
| 131 | save_path = os.path.join(self.working_dir, "story.txt") |
| 132 | if os.path.exists(save_path): |
| 133 | with open(save_path, "r", encoding="utf-8") as f: |
| 134 | story = f.read() |
| 135 | _pipeline_print(quiet, f"🚀 Loaded story from existing file.") |
| 136 | else: |
| 137 | _pipeline_print(quiet, "🧠 Developing story...") |
| 138 | story = await self.screenwriter.develop_story(idea=idea, user_requirement=user_requirement) |
| 139 | with open(save_path, "w", encoding="utf-8") as f: |
| 140 | f.write(story) |
| 141 | _pipeline_print(quiet, f"✅ Developed story and saved to {save_path}.") |
| 142 | |
| 143 | return story |
| 144 | |
| 145 | async def write_script_based_on_story( |
| 146 | self, |
| 147 | story: str, |
| 148 | user_requirement: str, |
| 149 | quiet: bool = False, |
| 150 | ): |
| 151 | save_path = os.path.join(self.working_dir, "script.json") |
| 152 | if os.path.exists(save_path): |
| 153 | with open(save_path, "r", encoding="utf-8") as f: |
| 154 | script = json.load(f) |
| 155 | _pipeline_print(quiet, f"🚀 Loaded script from existing file.") |
| 156 | else: |
| 157 | _pipeline_print(quiet, "🧠 Writing script based on story...") |
| 158 | script = await self.screenwriter.write_script_based_on_story(story=story, user_requirement=user_requirement) |
| 159 | with open(save_path, "w", encoding="utf-8") as f: |
| 160 | json.dump(script, f, ensure_ascii=False, indent=4) |
| 161 | _pipeline_print(quiet, f"✅ Written script based on story and saved to {save_path}.") |
| 162 | return script |
| 163 | |
| 164 | async def generate_portraits_for_single_character( |
| 165 | self, |
| 166 | character: CharacterInScene, |
| 167 | style: str, |
| 168 | ): |
| 169 | character_dir = os.path.join( |
| 170 | self.working_dir, "character_portraits", f"{character.idx}_{safe_path_component(character.identifier_in_scene)}") |
| 171 | os.makedirs(character_dir, exist_ok=True) |
| 172 | |
| 173 | front_portrait_path = os.path.join(character_dir, "front.png") |
| 174 | if os.path.exists(front_portrait_path): |
| 175 | pass |
| 176 | else: |
| 177 | front_portrait_output = await self.character_portraits_generator.generate_front_portrait(character, style) |
| 178 | front_portrait_output.save(front_portrait_path) |
| 179 | |
| 180 | side_portrait_path = os.path.join(character_dir, "side.png") |
| 181 | if os.path.exists(side_portrait_path): |
| 182 | pass |
| 183 | else: |
| 184 | try: |
| 185 | side_portrait_output = await self.character_portraits_generator.generate_side_portrait(character, front_portrait_path) |
| 186 | side_portrait_output.save(side_portrait_path) |
| 187 | except Exception as e: |
| 188 | # gemini-2.5-flash-image intermittently (sometimes beyond |
| 189 | # the tenacity retry budget) fails this front->side |
| 190 | # re-angling edit with finish_reason=IMAGE_OTHER / empty |
| 191 | # content. Fall back to the front portrait rather than |
| 192 | # aborting the whole pipeline. |
| 193 | print(f"⚠️ Side portrait generation failed for {character.identifier_in_scene} after retries ({e}); reusing front portrait as fallback.") |
| 194 | shutil.copy(front_portrait_path, side_portrait_path) |
| 195 | |
| 196 | back_portrait_path = os.path.join(character_dir, "back.png") |
| 197 | if os.path.exists(back_portrait_path): |
| 198 | pass |
| 199 | else: |
| 200 | try: |
| 201 | back_portrait_output = await self.character_portraits_generator.generate_back_portrait(character, front_portrait_path) |
| 202 | back_portrait_output.save(back_portrait_path) |
| 203 | except Exception as e: |
| 204 | print(f"⚠️ Back portrait generation failed for {character.identifier_in_scene} after retries ({e}); reusing front portrait as fallback.") |
| 205 | shutil.copy(front_portrait_path, back_portrait_path) |
| 206 | |
| 207 | print( |
| 208 | f"☑️ Completed character portrait generation for {character.identifier_in_scene}.") |
| 209 | |
| 210 | return { |
| 211 | character.identifier_in_scene: { |
| 212 | "front": { |
| 213 | "path": front_portrait_path, |
| 214 | "description": f"A front view portrait of {character.identifier_in_scene}.", |
| 215 | }, |
| 216 | "side": { |
| 217 | "path": side_portrait_path, |
| 218 | "description": f"A side view portrait of {character.identifier_in_scene}.", |
| 219 | }, |
| 220 | "back": { |
| 221 | "path": back_portrait_path, |
| 222 | "description": f"A back view portrait of {character.identifier_in_scene}.", |
| 223 | }, |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | async def __call__( |
| 228 | self, |
| 229 | idea: str, |
| 230 | user_requirement: str, |
| 231 | style: str, |
| 232 | quiet: bool = False, |
| 233 | progress: Callable[[str, str, Dict[str, Any] | None], None] | None = None, |
| 234 | ): |
| 235 | |
| 236 | story = await self.develop_story(idea=idea, user_requirement=user_requirement, quiet=quiet) |
| 237 | |
| 238 | characters = await self.extract_characters(story=story, quiet=quiet) |
| 239 | |
| 240 | character_portraits_registry = await self.generate_character_portraits( |
| 241 | characters=characters, |
| 242 | character_portraits_registry=None, |
| 243 | style=style, |
| 244 | ) |
| 245 | |
| 246 | scene_scripts = await self.write_script_based_on_story(story=story, user_requirement=user_requirement, quiet=quiet) |
| 247 | |
| 248 | all_video_paths = [] |
| 249 | |
| 250 | for idx, scene_script in enumerate(scene_scripts): |
| 251 | scene_working_dir = os.path.join(self.working_dir, f"scene_{idx}") |
| 252 | os.makedirs(scene_working_dir, exist_ok=True) |
| 253 | script2video_pipeline = Script2VideoPipeline( |
| 254 | chat_model=self.chat_model, |
| 255 | image_generator=self.image_generator, |
| 256 | video_generator=self.video_generator, |
| 257 | working_dir=scene_working_dir, |
| 258 | num_image_candidates=self.num_image_candidates, |
| 259 | ) |
| 260 | final_video_path = await script2video_pipeline( |
| 261 | script=scene_script, |
| 262 | user_requirement=user_requirement, |
| 263 | style=style, |
| 264 | characters=characters, |
| 265 | character_portraits_registry=character_portraits_registry, |
| 266 | quiet=quiet, |
| 267 | progress=_scoped_progress(progress, scene_idx=idx), |
| 268 | ) |
| 269 | all_video_paths.append(final_video_path) |
| 270 | |
| 271 | final_video_path = os.path.join(self.working_dir, "final_video.mp4") |
| 272 | if os.path.exists(final_video_path): |
| 273 | _pipeline_print(quiet, f"🚀 Skipped concatenating videos, already exists.") |
| 274 | else: |
| 275 | _pipeline_print(quiet, f"🎬 Starting concatenating videos...") |
| 276 | concatenate_video_files(all_video_paths, final_video_path) |
| 277 | _pipeline_print(quiet, f"☑️ Concatenated videos, saved to {final_video_path}.") |
| 278 | return final_video_path |
| 279 |