| 1 | import logging |
| 2 | from typing import List, Literal |
| 3 | import asyncio |
| 4 | import aiohttp |
| 5 | from interfaces.video_output import VideoOutput |
| 6 | from utils.image import image_path_to_b64 |
| 7 | |
| 8 | |
| 9 | class VideoGeneratorDoubaoSeedanceYunwuAPI: |
| 10 | def __init__( |
| 11 | self, |
| 12 | api_key: str, |
| 13 | t2v_model: str = "doubao-seedance-1-0-lite-t2v-250428", |
| 14 | ff2v_model: str = "doubao-seedance-1-0-lite-i2v-250428", |
| 15 | flf2v_model: str = "doubao-seedance-1-0-lite-i2v-250428", |
| 16 | max_create_attempts: int = 3, |
| 17 | poll_interval: int = 2, |
| 18 | max_poll_attempts: int = 300, |
| 19 | ): |
| 20 | self.api_key = api_key |
| 21 | self.t2v_model = t2v_model |
| 22 | self.ff2v_model = ff2v_model |
| 23 | self.flf2v_model = flf2v_model |
| 24 | self.max_create_attempts = max_create_attempts |
| 25 | self.poll_interval = poll_interval |
| 26 | self.max_poll_attempts = max_poll_attempts |
| 27 | |
| 28 | |
| 29 | async def create_video_generation_task( |
| 30 | self, |
| 31 | prompt: str, |
| 32 | reference_image_paths: List[str], |
| 33 | resolution: Literal["480p", "720p", "1080p"] = "720p", |
| 34 | aspect_ratio: str = "16:9", |
| 35 | fps: Literal[16, 24] = 16, |
| 36 | duration: Literal[5, 10] = 5, |
| 37 | ) -> str: |
| 38 | """ |
| 39 | Create a video generation task and return the task ID. |
| 40 | |
| 41 | Args: |
| 42 | prompt: Text prompt for video generation |
| 43 | reference_image_paths: List of 1 or 2 reference images |
| 44 | |
| 45 | Returns: |
| 46 | Task ID string |
| 47 | """ |
| 48 | if len(reference_image_paths) == 0: |
| 49 | model = self.t2v_model |
| 50 | elif len(reference_image_paths) == 1: |
| 51 | model = self.ff2v_model |
| 52 | elif len(reference_image_paths) == 2: |
| 53 | model = self.flf2v_model |
| 54 | else: |
| 55 | raise ValueError("reference_image_paths must contain 1 or 2 images.") |
| 56 | |
| 57 | logging.info(f"Calling {model} to generate video...") |
| 58 | |
| 59 | url = "https://yunwu.ai/volc/v1/contents/generations/tasks" |
| 60 | |
| 61 | |
| 62 | content = [ |
| 63 | { |
| 64 | "type": "text", |
| 65 | "text": prompt + f" --rs {resolution} --rt {aspect_ratio} --dur {duration} --fps {fps} --wm false --seed -1 --cf false" |
| 66 | } |
| 67 | ] |
| 68 | if len(reference_image_paths) >= 1: |
| 69 | content.append( |
| 70 | { |
| 71 | "type": "image_url", |
| 72 | "image_url": { |
| 73 | "url": image_path_to_b64(reference_image_paths[0]) |
| 74 | }, |
| 75 | "role": "first_frame", |
| 76 | } |
| 77 | ) |
| 78 | if len(reference_image_paths) >= 2: |
| 79 | content.append( |
| 80 | { |
| 81 | "type": "image_url", |
| 82 | "image_url": { |
| 83 | "url": image_path_to_b64(reference_image_paths[1]) |
| 84 | }, |
| 85 | "role": "last_frame", |
| 86 | } |
| 87 | ) |
| 88 | |
| 89 | payload = { |
| 90 | "model": model, |
| 91 | "content": content |
| 92 | } |
| 93 | |
| 94 | headers = { |
| 95 | 'Authorization': f'Bearer {self.api_key}', |
| 96 | 'Content-Type': 'application/json' |
| 97 | } |
| 98 | |
| 99 | last_error = None |
| 100 | for attempt in range(1, self.max_create_attempts + 1): |
| 101 | try: |
| 102 | async with aiohttp.ClientSession() as session: |
| 103 | async with session.post(url, headers=headers, json=payload) as response: |
| 104 | response_json = await response.json() |
| 105 | http_status = response.status |
| 106 | logging.debug(f"Response: {response_json}") |
| 107 | except Exception as e: |
| 108 | last_error = e |
| 109 | logging.error(f"Error occurred while creating video generation task (attempt {attempt}/{self.max_create_attempts}): {e}") |
| 110 | if attempt < self.max_create_attempts: |
| 111 | await asyncio.sleep(attempt) |
| 112 | continue |
| 113 | |
| 114 | if http_status >= 400: |
| 115 | message = f"Video generation task creation failed with HTTP {http_status}: {response_json}" |
| 116 | if http_status < 500: |
| 117 | raise RuntimeError(message) |
| 118 | last_error = RuntimeError(message) |
| 119 | logging.error(f"{message} (attempt {attempt}/{self.max_create_attempts})") |
| 120 | if attempt < self.max_create_attempts: |
| 121 | await asyncio.sleep(attempt) |
| 122 | continue |
| 123 | |
| 124 | task_id = response_json.get("id") |
| 125 | if not task_id: |
| 126 | raise RuntimeError(f"Video generation task creation returned no task id: {response_json}") |
| 127 | logging.info(f"Video generation task created successfully. Task ID: {task_id}") |
| 128 | return task_id |
| 129 | |
| 130 | raise RuntimeError(f"Failed to create video generation task after {self.max_create_attempts} attempts.") from last_error |
| 131 | |
| 132 | async def query_video_generation_task( |
| 133 | self, |
| 134 | task_id: str, |
| 135 | ) -> str: |
| 136 | """ |
| 137 | Query the video generation task until completion and return the video URL. |
| 138 | |
| 139 | Args: |
| 140 | task_id: Task ID to query |
| 141 | |
| 142 | Returns: |
| 143 | Video URL string |
| 144 | """ |
| 145 | url = f"https://yunwu.ai/volc/v1/contents/generations/tasks/{task_id}" |
| 146 | headers = { |
| 147 | 'Authorization': f'Bearer {self.api_key}', |
| 148 | } |
| 149 | |
| 150 | attempts = 0 |
| 151 | consecutive_errors = 0 |
| 152 | while True: |
| 153 | if attempts >= self.max_poll_attempts: |
| 154 | raise TimeoutError(f"Video generation did not complete after {attempts} polls.") |
| 155 | attempts += 1 |
| 156 | |
| 157 | try: |
| 158 | async with aiohttp.ClientSession() as session: |
| 159 | async with session.get(url, headers=headers) as response: |
| 160 | response_json = await response.json() |
| 161 | http_status = response.status |
| 162 | except Exception as e: |
| 163 | consecutive_errors += 1 |
| 164 | if consecutive_errors >= 5: |
| 165 | raise RuntimeError(f"Querying video generation task failed {consecutive_errors} times in a row.") from e |
| 166 | logging.error(f"Error occurred while querying video generation task: {e}. Retrying in {self.poll_interval} seconds...") |
| 167 | await asyncio.sleep(self.poll_interval) |
| 168 | continue |
| 169 | consecutive_errors = 0 |
| 170 | |
| 171 | if http_status >= 400: |
| 172 | raise RuntimeError(f"Querying video generation task failed with HTTP {http_status}: {response_json}") |
| 173 | |
| 174 | status = response_json.get("status") |
| 175 | if status == "succeeded": |
| 176 | video_url = response_json["content"]["video_url"] |
| 177 | logging.info(f"Video generation completed successfully. Video URL: {video_url}") |
| 178 | return video_url |
| 179 | elif status == "failed": |
| 180 | logging.error(f"Video generation failed. Response: {response_json}") |
| 181 | raise ValueError("Video generation failed.") |
| 182 | else: |
| 183 | logging.info(f"Video generation is still in progress. Checking again in {self.poll_interval} seconds...") |
| 184 | await asyncio.sleep(self.poll_interval) |
| 185 | |
| 186 | async def generate_single_video( |
| 187 | self, |
| 188 | prompt: str, |
| 189 | reference_image_paths: List[str], |
| 190 | resolution: Literal["480p", "720p", "1080p"] = "720p", |
| 191 | aspect_ratio: str = "16:9", |
| 192 | fps: Literal[16, 24] = 16, |
| 193 | duration: Literal[5, 10] = 5, |
| 194 | **kwargs, |
| 195 | ) -> VideoOutput: |
| 196 | """ |
| 197 | Generate a single video by creating a task and waiting for completion. |
| 198 | |
| 199 | Args: |
| 200 | prompt: Text prompt for video generation |
| 201 | reference_image_paths: List of 1 or 2 reference images |
| 202 | resolution: Resolution of the video |
| 203 | aspect_ratio: Aspect ratio of the video |
| 204 | fps: Frames per second of the video |
| 205 | duration: Duration of the video |
| 206 | Returns: |
| 207 | VideoOutput containing the video URL |
| 208 | """ |
| 209 | task_id = await self.create_video_generation_task(prompt, reference_image_paths, resolution, aspect_ratio, fps, duration) |
| 210 | video_url = await self.query_video_generation_task(task_id) |
| 211 | return VideoOutput(fmt="url", ext="mp4", data=video_url) |
| 212 | |
| 213 |