| 1 | #!/usr/bin/env python3 |
| 2 | """ |
| 3 | MiniMax image generation backend. |
| 4 | |
| 5 | Configuration keys: |
| 6 | MINIMAX_API_KEY (required) |
| 7 | MINIMAX_BASE_URL (optional) |
| 8 | MINIMAX_MODEL (optional; image-01 only) |
| 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 minimax") |
| 25 | raise SystemExit(0 if any(arg in {"-h", "--help", "help"} for arg in sys.argv[1:]) else 1) |
| 26 | |
| 27 | import base64 |
| 28 | import os |
| 29 | import time |
| 30 | |
| 31 | import requests |
| 32 | |
| 33 | from image_backends.backend_common import ( |
| 34 | MAX_RETRIES, |
| 35 | detect_image_extension, |
| 36 | download_image, |
| 37 | http_error, |
| 38 | is_permanent_error, |
| 39 | is_rate_limit_error, |
| 40 | normalize_image_size, |
| 41 | require_api_key, |
| 42 | resolve_output_path, |
| 43 | retry_delay, |
| 44 | save_image_bytes, |
| 45 | ) |
| 46 | |
| 47 | |
| 48 | DEFAULT_ENDPOINT = "https://api.minimaxi.com/v1/image_generation" |
| 49 | DEFAULT_MODEL = "image-01" |
| 50 | SUPPORTED_MODELS = {DEFAULT_MODEL} |
| 51 | |
| 52 | # Request the documented default response format. The API returns hosted links |
| 53 | # in `data.image_urls` for "url" and inline strings in `data.image_base64` for |
| 54 | # "base64"; both shapes are handled when the response is parsed. |
| 55 | DEFAULT_RESPONSE_FORMAT = "url" |
| 56 | |
| 57 | # International fallback: set MINIMAX_BASE_URL=https://api.minimax.io if needed |
| 58 | |
| 59 | ASPECT_RATIO_SIZE_MAP = { |
| 60 | "512px": { |
| 61 | "1:1": (512, 512), |
| 62 | "16:9": (912, 512), |
| 63 | "4:3": (680, 512), |
| 64 | "3:2": (768, 512), |
| 65 | "2:3": (512, 768), |
| 66 | "3:4": (512, 680), |
| 67 | "9:16": (512, 912), |
| 68 | "21:9": (1192, 512), |
| 69 | }, |
| 70 | "1K": { |
| 71 | "1:1": (1024, 1024), |
| 72 | "16:9": (1280, 720), |
| 73 | "4:3": (1152, 864), |
| 74 | "3:2": (1248, 832), |
| 75 | "2:3": (832, 1248), |
| 76 | "3:4": (864, 1152), |
| 77 | "9:16": (720, 1280), |
| 78 | "21:9": (1344, 576), |
| 79 | }, |
| 80 | "2K": { |
| 81 | "1:1": (2048, 2048), |
| 82 | "16:9": (2048, 1152), |
| 83 | "4:3": (2048, 1536), |
| 84 | "3:2": (2048, 1368), |
| 85 | "2:3": (1368, 2048), |
| 86 | "3:4": (1536, 2048), |
| 87 | "9:16": (1152, 2048), |
| 88 | "21:9": (2048, 880), |
| 89 | }, |
| 90 | } |
| 91 | |
| 92 | |
| 93 | def _validate_model(model: str) -> str: |
| 94 | """Limit the backend to the model contract implemented below.""" |
| 95 | resolved = model.strip() |
| 96 | if resolved not in SUPPORTED_MODELS: |
| 97 | raise ValueError( |
| 98 | f"Unsupported MiniMax model '{model}'. Supported: {sorted(SUPPORTED_MODELS)}" |
| 99 | ) |
| 100 | return resolved |
| 101 | |
| 102 | |
| 103 | def _resolve_url(base_url: str) -> str: |
| 104 | """Resolve the MiniMax image generation endpoint. |
| 105 | |
| 106 | Accepts three forms of MINIMAX_BASE_URL: |
| 107 | - Full endpoint: https://api.minimax.io/v1/image_generation → used as-is |
| 108 | - Versioned base: https://api.minimax.io/v1 → appends /image_generation |
| 109 | - Root base: https://api.minimax.io → appends /v1/image_generation |
| 110 | """ |
| 111 | base = base_url.rstrip("/") |
| 112 | if base.endswith("/image_generation"): |
| 113 | return base |
| 114 | if base.endswith("/v1"): |
| 115 | return base + "/image_generation" |
| 116 | return base + "/v1/image_generation" |
| 117 | |
| 118 | |
| 119 | def _resolve_dimensions(aspect_ratio: str, image_size: str) -> tuple[int, int]: |
| 120 | """Resolve width and height from the unified aspect_ratio/image_size pair.""" |
| 121 | normalized = normalize_image_size(image_size) |
| 122 | sizes = ASPECT_RATIO_SIZE_MAP.get(normalized) |
| 123 | if sizes is None: |
| 124 | supported_sizes = ", ".join(ASPECT_RATIO_SIZE_MAP) |
| 125 | raise ValueError( |
| 126 | f"Unsupported image size '{image_size}' for MiniMax backend. " |
| 127 | f"image-01 supports these logical sizes: {supported_sizes}." |
| 128 | ) |
| 129 | dimensions = sizes.get(aspect_ratio) |
| 130 | if not dimensions: |
| 131 | supported = sorted(sizes) |
| 132 | raise ValueError( |
| 133 | f"Unsupported aspect ratio '{aspect_ratio}' for MiniMax backend. " |
| 134 | f"Supported: {supported}" |
| 135 | ) |
| 136 | return dimensions |
| 137 | |
| 138 | |
| 139 | def _extract_image_bytes(payload: dict) -> bytes | None: |
| 140 | """Extract inline base64 image bytes from a MiniMax response payload.""" |
| 141 | data = payload.get("data") or {} |
| 142 | image_base64 = data.get("image_base64") or [] |
| 143 | if image_base64: |
| 144 | return base64.b64decode(image_base64[0]) |
| 145 | return None |
| 146 | |
| 147 | |
| 148 | def _extract_image_url(payload: dict) -> str | None: |
| 149 | """Extract the first hosted image URL from a MiniMax response payload.""" |
| 150 | data = payload.get("data") or {} |
| 151 | image_urls = data.get("image_urls") or [] |
| 152 | if image_urls: |
| 153 | return image_urls[0] |
| 154 | return None |
| 155 | |
| 156 | |
| 157 | def _generate_image(api_key: str, prompt: str, |
| 158 | aspect_ratio: str = "1:1", image_size: str = "1K", |
| 159 | output_dir: str = None, filename: str = None, |
| 160 | model: str = DEFAULT_MODEL, base_url: str = DEFAULT_ENDPOINT) -> str: |
| 161 | """Generate one image with the MiniMax backend.""" |
| 162 | model = _validate_model(model) |
| 163 | width, height = _resolve_dimensions(aspect_ratio, image_size) |
| 164 | url = _resolve_url(base_url) |
| 165 | |
| 166 | headers = { |
| 167 | "Authorization": f"Bearer {api_key}", |
| 168 | "Content-Type": "application/json", |
| 169 | } |
| 170 | payload = { |
| 171 | "model": model, |
| 172 | "prompt": prompt, |
| 173 | "width": width, |
| 174 | "height": height, |
| 175 | "response_format": DEFAULT_RESPONSE_FORMAT, |
| 176 | "n": 1, |
| 177 | } |
| 178 | |
| 179 | print("[MiniMax Image]") |
| 180 | print(f" Model: {model}") |
| 181 | print(f" Prompt: {prompt[:120]}{'...' if len(prompt) > 120 else ''}") |
| 182 | print(f" Aspect Ratio: {aspect_ratio}") |
| 183 | print(f" Resolution: {width}x{height} (from image_size={image_size})") |
| 184 | print() |
| 185 | print(" [..] Generating...", end="", flush=True) |
| 186 | start = time.time() |
| 187 | response = requests.post(url, headers=headers, json=payload, timeout=300) |
| 188 | elapsed = time.time() - start |
| 189 | print(f"\n [DONE] Response received ({elapsed:.1f}s)") |
| 190 | |
| 191 | if response.status_code != 200: |
| 192 | raise http_error(response, "MiniMax image generation") |
| 193 | |
| 194 | data = response.json() |
| 195 | base_resp = data.get("base_resp") or {} |
| 196 | if base_resp.get("status_code") not in (None, 0, "0"): |
| 197 | raise RuntimeError(f"MiniMax image generation failed: {data}") |
| 198 | |
| 199 | image_bytes = _extract_image_bytes(data) |
| 200 | if image_bytes: |
| 201 | ext = detect_image_extension(image_bytes) or ".jpeg" |
| 202 | path = resolve_output_path(prompt, output_dir, filename, ext) |
| 203 | return save_image_bytes(image_bytes, path) |
| 204 | |
| 205 | image_url = _extract_image_url(data) |
| 206 | if image_url: |
| 207 | # image-01 serves JPEG; save_image_bytes realigns the extension when the |
| 208 | # downloaded bytes are a different format. |
| 209 | path = resolve_output_path(prompt, output_dir, filename, ".jpeg") |
| 210 | return download_image(image_url, path) |
| 211 | |
| 212 | raise RuntimeError(f"MiniMax response missing image data: {data}") |
| 213 | |
| 214 | |
| 215 | def generate(prompt: str, |
| 216 | aspect_ratio: str = "1:1", image_size: str = "1K", |
| 217 | output_dir: str = None, filename: str = None, |
| 218 | model: str = None, max_retries: int = MAX_RETRIES) -> str: |
| 219 | """Generate an image with retries using the MiniMax backend.""" |
| 220 | resolved_model = model or os.environ.get("MINIMAX_MODEL") or DEFAULT_MODEL |
| 221 | _validate_model(resolved_model) |
| 222 | normalized_size = normalize_image_size(image_size) |
| 223 | _resolve_dimensions(aspect_ratio, normalized_size) |
| 224 | api_key = require_api_key( |
| 225 | "MINIMAX_API_KEY", |
| 226 | message="No API key found. Set MINIMAX_API_KEY in the current environment or a .env file.", |
| 227 | ) |
| 228 | base_url = os.environ.get("MINIMAX_BASE_URL") or DEFAULT_ENDPOINT |
| 229 | |
| 230 | last_error = None |
| 231 | for attempt in range(max_retries + 1): |
| 232 | try: |
| 233 | return _generate_image( |
| 234 | api_key=api_key, |
| 235 | prompt=prompt, |
| 236 | aspect_ratio=aspect_ratio, |
| 237 | image_size=normalized_size, |
| 238 | output_dir=output_dir, |
| 239 | filename=filename, |
| 240 | model=resolved_model, |
| 241 | base_url=base_url, |
| 242 | ) |
| 243 | except Exception as exc: |
| 244 | last_error = exc |
| 245 | if is_permanent_error(exc): |
| 246 | raise |
| 247 | if attempt >= max_retries: |
| 248 | break |
| 249 | limited = is_rate_limit_error(exc) |
| 250 | delay = retry_delay(attempt, rate_limited=limited) |
| 251 | label = "Rate limit hit" if limited else f"Error: {exc}" |
| 252 | print(f"\n [WARN] {label}. Retrying in {delay}s...") |
| 253 | time.sleep(delay) |
| 254 | |
| 255 | raise RuntimeError(f"Failed after {max_retries + 1} attempts. Last error: {last_error}") |
| 256 |