| 1 | #!/usr/bin/env python3 |
| 2 | """ |
| 3 | Volcengine Seedream image generation backend. |
| 4 | |
| 5 | Configuration keys: |
| 6 | VOLCENGINE_API_KEY / ARK_API_KEY (required) |
| 7 | VOLCENGINE_BASE_URL (optional) |
| 8 | VOLCENGINE_MODEL (optional) |
| 9 | """ |
| 10 | |
| 11 | import sys |
| 12 | from pathlib import Path |
| 13 | |
| 14 | _SCRIPTS_DIR = Path(__file__).resolve().parents[1] |
| 15 | if str(_SCRIPTS_DIR) not in sys.path: |
| 16 | sys.path.insert(0, str(_SCRIPTS_DIR)) |
| 17 | |
| 18 | from console_encoding import configure_utf8_stdio # noqa: E402 |
| 19 | |
| 20 | configure_utf8_stdio() |
| 21 | |
| 22 | if __name__ == "__main__": |
| 23 | print(__doc__) |
| 24 | print("Use via: python3 skills/ppt-master/scripts/image_gen.py \"prompt\" --backend volcengine") |
| 25 | raise SystemExit(0 if any(arg in {"-h", "--help", "help"} for arg in sys.argv[1:]) else 1) |
| 26 | |
| 27 | import os |
| 28 | import time |
| 29 | |
| 30 | import requests |
| 31 | |
| 32 | from image_backends.backend_common import ( |
| 33 | MAX_RETRIES, |
| 34 | download_image, |
| 35 | http_error, |
| 36 | is_rate_limit_error, |
| 37 | normalize_image_size, |
| 38 | require_api_key, |
| 39 | resolve_output_path, |
| 40 | retry_delay, |
| 41 | ) |
| 42 | |
| 43 | |
| 44 | DEFAULT_ENDPOINT = "https://operator.las.cn-beijing.volces.com/api/v1/images/generations" |
| 45 | DEFAULT_MODEL = "doubao-seedream-4-5-251128" |
| 46 | |
| 47 | ASPECT_RATIO_SIZE_MAP = { |
| 48 | "512px": { |
| 49 | "1:1": "1024x1024", |
| 50 | "2:3": "1024x1536", |
| 51 | "3:2": "1536x1024", |
| 52 | "3:4": "1024x1365", |
| 53 | "4:3": "1365x1024", |
| 54 | "4:5": "1024x1280", |
| 55 | "5:4": "1280x1024", |
| 56 | "9:16": "1024x1820", |
| 57 | "16:9": "1820x1024", |
| 58 | "21:9": "2048x878", |
| 59 | }, |
| 60 | "1K": { |
| 61 | "1:1": "1536x1536", |
| 62 | "2:3": "1344x2016", |
| 63 | "3:2": "2016x1344", |
| 64 | "3:4": "1440x1920", |
| 65 | "4:3": "1920x1440", |
| 66 | "4:5": "1536x1920", |
| 67 | "5:4": "1920x1536", |
| 68 | "9:16": "1152x2048", |
| 69 | "16:9": "2048x1152", |
| 70 | "21:9": "2048x878", |
| 71 | }, |
| 72 | "2K": { |
| 73 | "1:1": "2048x2048", |
| 74 | "2:3": "1536x2048", |
| 75 | "3:2": "2048x1536", |
| 76 | "3:4": "1536x2048", |
| 77 | "4:3": "2048x1536", |
| 78 | "4:5": "1638x2048", |
| 79 | "5:4": "2048x1638", |
| 80 | "9:16": "1152x2048", |
| 81 | "16:9": "2048x1152", |
| 82 | "21:9": "2048x878", |
| 83 | }, |
| 84 | "4K": { |
| 85 | "1:1": "2048x2048", |
| 86 | "2:3": "1536x2048", |
| 87 | "3:2": "2048x1536", |
| 88 | "3:4": "1536x2048", |
| 89 | "4:3": "2048x1536", |
| 90 | "4:5": "1638x2048", |
| 91 | "5:4": "2048x1638", |
| 92 | "9:16": "1152x2048", |
| 93 | "16:9": "2048x1152", |
| 94 | "21:9": "2048x878", |
| 95 | }, |
| 96 | } |
| 97 | |
| 98 | |
| 99 | def _resolve_url(base_url: str) -> str: |
| 100 | """Resolve the Volcengine generation endpoint.""" |
| 101 | base = base_url.rstrip("/") |
| 102 | if base.endswith("/images/generations"): |
| 103 | return base |
| 104 | return base + "/api/v1/images/generations" |
| 105 | |
| 106 | |
| 107 | def _resolve_size(aspect_ratio: str, image_size: str) -> str: |
| 108 | """Resolve the target resolution for a ratio and logical size preset.""" |
| 109 | normalized = normalize_image_size(image_size) |
| 110 | size = (ASPECT_RATIO_SIZE_MAP.get(normalized) or {}).get(aspect_ratio) |
| 111 | if not size: |
| 112 | supported = sorted(ASPECT_RATIO_SIZE_MAP["1K"]) |
| 113 | raise ValueError( |
| 114 | f"Unsupported aspect ratio '{aspect_ratio}' for Volcengine backend. " |
| 115 | f"Supported: {supported}" |
| 116 | ) |
| 117 | return size |
| 118 | |
| 119 | |
| 120 | def _generate_image(api_key: str, prompt: str, |
| 121 | aspect_ratio: str = "1:1", image_size: str = "1K", |
| 122 | output_dir: str = None, filename: str = None, |
| 123 | model: str = DEFAULT_MODEL, base_url: str = DEFAULT_ENDPOINT) -> str: |
| 124 | """Generate one image with the Volcengine backend.""" |
| 125 | size = _resolve_size(aspect_ratio, image_size) |
| 126 | url = _resolve_url(base_url) |
| 127 | headers = { |
| 128 | "Authorization": f"Bearer {api_key}", |
| 129 | "Content-Type": "application/json", |
| 130 | } |
| 131 | payload = { |
| 132 | "model": model, |
| 133 | "prompt": prompt, |
| 134 | "size": size, |
| 135 | "response_format": "url", |
| 136 | "watermark": False, |
| 137 | } |
| 138 | |
| 139 | print("[Volcengine Seedream]") |
| 140 | print(f" Model: {model}") |
| 141 | print(f" Prompt: {prompt[:120]}{'...' if len(prompt) > 120 else ''}") |
| 142 | print(f" Aspect Ratio: {aspect_ratio}") |
| 143 | print(f" Resolution: {size}") |
| 144 | print() |
| 145 | print(" [..] Generating...", end="", flush=True) |
| 146 | start = time.time() |
| 147 | response = requests.post(url, headers=headers, json=payload, timeout=300) |
| 148 | elapsed = time.time() - start |
| 149 | print(f"\n [DONE] Response received ({elapsed:.1f}s)") |
| 150 | |
| 151 | if response.status_code != 200: |
| 152 | raise http_error(response, "Volcengine image generation") |
| 153 | |
| 154 | data = response.json() |
| 155 | items = data.get("data") or [] |
| 156 | image_url = items[0].get("url") if items else None |
| 157 | if not image_url: |
| 158 | raise RuntimeError(f"Volcengine response missing image URL: {data}") |
| 159 | |
| 160 | path = resolve_output_path(prompt, output_dir, filename, ".jpeg") |
| 161 | return download_image(image_url, path) |
| 162 | |
| 163 | |
| 164 | def generate(prompt: str, |
| 165 | aspect_ratio: str = "1:1", image_size: str = "1K", |
| 166 | output_dir: str = None, filename: str = None, |
| 167 | model: str = None, max_retries: int = MAX_RETRIES) -> str: |
| 168 | """Generate an image with retries using the Volcengine backend.""" |
| 169 | api_key = require_api_key( |
| 170 | "VOLCENGINE_API_KEY", |
| 171 | "ARK_API_KEY", |
| 172 | message="No API key found. Set VOLCENGINE_API_KEY or ARK_API_KEY in the current environment or a .env file.", |
| 173 | ) |
| 174 | base_url = os.environ.get("VOLCENGINE_BASE_URL") or DEFAULT_ENDPOINT |
| 175 | resolved_model = model or os.environ.get("VOLCENGINE_MODEL") or DEFAULT_MODEL |
| 176 | |
| 177 | last_error = None |
| 178 | for attempt in range(max_retries + 1): |
| 179 | try: |
| 180 | return _generate_image( |
| 181 | api_key=api_key, |
| 182 | prompt=prompt, |
| 183 | aspect_ratio=aspect_ratio, |
| 184 | image_size=image_size, |
| 185 | output_dir=output_dir, |
| 186 | filename=filename, |
| 187 | model=resolved_model, |
| 188 | base_url=base_url, |
| 189 | ) |
| 190 | except Exception as exc: |
| 191 | last_error = exc |
| 192 | if attempt >= max_retries: |
| 193 | break |
| 194 | limited = is_rate_limit_error(exc) |
| 195 | delay = retry_delay(attempt, rate_limited=limited) |
| 196 | label = "Rate limit hit" if limited else f"Error: {exc}" |
| 197 | print(f"\n [WARN] {label}. Retrying in {delay}s...") |
| 198 | time.sleep(delay) |
| 199 | |
| 200 | raise RuntimeError(f"Failed after {max_retries + 1} attempts. Last error: {last_error}") |
| 201 |