Skip to content

AI turns documents or topics into real, native PowerPoint decks—with native shapes, transitions and animations, data-backed charts and tables on demand, audio narration from speaker notes, and support for your own .pptx templates. · by Hugo He

fix(openai): support gpt-image-2 while preserving legacy compatibility

hugohe3a77499b25db0
diff --git a/.env.example b/.env.example
index 920c6f88..a3c4a55f 100644
--- a/.env.example
+++ b/.env.example
@@ -35,7 +35,7 @@
# ─────────────────────────────────────────────────────────────
# IMAGE_BACKEND=openai
# OPENAI_API_KEY=sk-xxx
-# OPENAI_MODEL=gpt-image-1
+# OPENAI_MODEL=gpt-image-2
# OPENAI_BASE_URL=http://127.0.0.1:3000/v1
# ─────────────────────────────────────────────────────────────
diff --git a/skills/ppt-master/scripts/image_backends/backend_openai.py b/skills/ppt-master/scripts/image_backends/backend_openai.py
index fd1ce4f7..57d8a84f 100644
--- a/skills/ppt-master/scripts/image_backends/backend_openai.py
+++ b/skills/ppt-master/scripts/image_backends/backend_openai.py
@@ -8,7 +8,7 @@ Used by image_gen.py as a backend module.
Configuration keys:
OPENAI_API_KEY (required) API key
OPENAI_BASE_URL (optional) Custom API endpoint (e.g. http://127.0.0.1:3000/v1)
- OPENAI_MODEL (optional) Model name (default: gpt-image-1)
+ OPENAI_MODEL (optional) Model name (default: gpt-image-2)
Dependencies:
pip install openai Pillow
@@ -18,10 +18,12 @@ import base64
import os
import time
import threading
+from collections.abc import Mapping
from openai import OpenAI
from image_backends.backend_common import (
MAX_RETRIES,
+ download_image,
is_rate_limit_error,
normalize_image_size,
resolve_output_path,
@@ -34,9 +36,9 @@ from image_backends.backend_common import (
# ║ Constants ║
# ╚══════════════════════════════════════════════════════════════════╝
-# Aspect ratio -> OpenAI size mapping
-# Covers common PPT/social media ratios
-ASPECT_RATIO_TO_SIZE = {
+# Aspect ratio -> DALL-E 3 / legacy compatible size mapping.
+# Unknown OpenAI-compatible models use this table to preserve old behavior.
+LEGACY_COMPAT_ASPECT_RATIO_TO_SIZE = {
"1:1": "1024x1024",
"16:9": "1792x1024",
"9:16": "1024x1792",
@@ -49,7 +51,57 @@ ASPECT_RATIO_TO_SIZE = {
"21:9": "1792x1024", # closest wide format
}
-VALID_ASPECT_RATIOS = list(ASPECT_RATIO_TO_SIZE.keys())
+# GPT Image 1/1.5/mini officially support only square, landscape, portrait, or auto.
+GPT_IMAGE_LEGACY_ASPECT_RATIO_TO_SIZE = {
+ "1:1": "1024x1024",
+ "16:9": "1536x1024",
+ "9:16": "1024x1536",
+ "3:2": "1536x1024",
+ "2:3": "1024x1536",
+ "4:3": "1536x1024",
+ "3:4": "1024x1536",
+ "4:5": "1024x1536",
+ "5:4": "1536x1024",
+ "21:9": "1536x1024",
+}
+
+# GPT Image 2 supports flexible sizes when both edges are multiples of 16,
+# the edge ratio is <= 3:1, and the total pixels are within model limits.
+GPT_IMAGE_2_SIZES = {
+ "512px": {
+ "1:1": "1024x1024", "16:9": "1152x648", "9:16": "648x1152",
+ "3:2": "1248x832", "2:3": "832x1248", "4:3": "1024x768",
+ "3:4": "768x1024", "4:5": "896x1120", "5:4": "1120x896",
+ "21:9": "1280x544",
+ },
+ "1K": {
+ "1:1": "1024x1024", "16:9": "1152x648", "9:16": "648x1152",
+ "3:2": "1248x832", "2:3": "832x1248", "4:3": "1024x768",
+ "3:4": "768x1024", "4:5": "896x1120", "5:4": "1120x896",
+ "21:9": "1280x544",
+ },
+ "2K": {
+ "1:1": "2048x2048", "16:9": "2048x1152", "9:16": "1152x2048",
+ "3:2": "2016x1344", "2:3": "1344x2016", "4:3": "1920x1440",
+ "3:4": "1440x1920", "4:5": "1600x2000", "5:4": "2000x1600",
+ "21:9": "2560x1088",
+ },
+ "4K": {
+ "1:1": "2880x2880", "16:9": "3840x2160", "9:16": "2160x3840",
+ "3:2": "3520x2352", "2:3": "2352x3520", "4:3": "3264x2448",
+ "3:4": "2448x3264", "4:5": "2560x3200", "5:4": "3200x2560",
+ "21:9": "3840x1648",
+ },
+}
+
+DALL_E_2_SIZE_BY_IMAGE_SIZE = {
+ "512px": "512x512",
+ "1K": "1024x1024",
+ "2K": "1024x1024",
+ "4K": "1024x1024",
+}
+
+VALID_ASPECT_RATIOS = list(LEGACY_COMPAT_ASPECT_RATIO_TO_SIZE.keys())
# image_size -> quality mapping
IMAGE_SIZE_TO_QUALITY = {
@@ -59,7 +111,46 @@ IMAGE_SIZE_TO_QUALITY = {
"4K": "high",
}
-DEFAULT_MODEL = "gpt-image-1"
+DEFAULT_MODEL = "gpt-image-2"
+
+
+def _field(value, name: str):
+ """Read a response field from either an SDK object or a dict."""
+ if isinstance(value, Mapping):
+ return value.get(name)
+ return getattr(value, name, None)
+
+
+def _normalized_model(model: str) -> str:
+ return (model or "").strip().lower()
+
+
+def _is_gpt_image_model(model: str) -> bool:
+ return _normalized_model(model).startswith("gpt-image-")
+
+
+def _is_gpt_image_2(model: str) -> bool:
+ return _normalized_model(model).startswith("gpt-image-2")
+
+
+def _is_dall_e_2(model: str) -> bool:
+ return _normalized_model(model) == "dall-e-2"
+
+
+def _select_size(model: str, aspect_ratio: str, image_size: str) -> str:
+ """Select a model-compatible size while preserving legacy fallbacks."""
+ if _is_gpt_image_2(model):
+ return GPT_IMAGE_2_SIZES[image_size][aspect_ratio]
+ if _is_gpt_image_model(model):
+ return GPT_IMAGE_LEGACY_ASPECT_RATIO_TO_SIZE[aspect_ratio]
+ if _is_dall_e_2(model):
+ return DALL_E_2_SIZE_BY_IMAGE_SIZE[image_size]
+ return LEGACY_COMPAT_ASPECT_RATIO_TO_SIZE[aspect_ratio]
+
+
+def _supports_response_format(model: str) -> bool:
+ """GPT Image models always return base64; DALL-E/compatible models may need this."""
+ return not _is_gpt_image_model(model)
# ╔══════════════════════════════════════════════════════════════════╗
@@ -89,7 +180,7 @@ def _generate_image(api_key: str, prompt: str, negative_prompt: str = None,
final_prompt += f"\n\nAvoid the following: {negative_prompt}"
# Map parameters
- size = ASPECT_RATIO_TO_SIZE.get(aspect_ratio, "1024x1024")
+ size = _select_size(model, aspect_ratio, image_size)
quality = IMAGE_SIZE_TO_QUALITY.get(image_size, "auto")
mode_label = f"Proxy: {base_url}" if base_url else "OpenAI API"
@@ -117,14 +208,17 @@ def _generate_image(api_key: str, prompt: str, negative_prompt: str = None,
hb_thread.start()
try:
- resp = client.images.generate(
- prompt=final_prompt,
- model=model,
- size=size,
- quality=quality,
- n=1,
- response_format="b64_json",
- )
+ request = {
+ "prompt": final_prompt,
+ "model": model,
+ "size": size,
+ "quality": quality,
+ "n": 1,
+ }
+ if _supports_response_format(model):
+ request["response_format"] = "b64_json"
+
+ resp = client.images.generate(**request)
finally:
heartbeat_stop.set()
hb_thread.join(timeout=1)
@@ -132,10 +226,17 @@ def _generate_image(api_key: str, prompt: str, negative_prompt: str = None,
elapsed = time.time() - start_time
print(f"\n [DONE] Image generated ({elapsed:.1f}s)")
- if resp is not None and resp.data:
+ data = _field(resp, "data") if resp is not None else None
+ if data:
path = resolve_output_path(prompt, output_dir, filename, ".png")
- image_data = base64.b64decode(resp.data[0].b64_json)
- return save_image_bytes(image_data, path)
+ first_image = data[0]
+ b64_json = _field(first_image, "b64_json")
+ image_url = _field(first_image, "url")
+ if b64_json:
+ image_data = base64.b64decode(b64_json)
+ return save_image_bytes(image_data, path)
+ if image_url:
+ return download_image(image_url, path)
raise RuntimeError("No image was generated. The server may have refused the request.")
@@ -163,7 +264,7 @@ def generate(prompt: str, negative_prompt: str = None,
image_size: Image size, mapped to OpenAI quality
output_dir: Output directory
filename: Output filename (without extension)
- model: Model name (default: gpt-image-1)
+ model: Model name (default: gpt-image-2)
max_retries: Maximum number of retries
Returns:
@@ -182,8 +283,8 @@ def generate(prompt: str, negative_prompt: str = None,
image_size = normalize_image_size(image_size)
- if aspect_ratio not in ASPECT_RATIO_TO_SIZE:
- supported = list(ASPECT_RATIO_TO_SIZE.keys())
+ if aspect_ratio not in LEGACY_COMPAT_ASPECT_RATIO_TO_SIZE:
+ supported = list(LEGACY_COMPAT_ASPECT_RATIO_TO_SIZE.keys())
raise ValueError(
f"Unsupported aspect ratio '{aspect_ratio}' for OpenAI backend. "
f"Supported: {supported}"
diff --git a/skills/ppt-master/scripts/image_gen.py b/skills/ppt-master/scripts/image_gen.py
index e2213ce6..f2258419 100644
--- a/skills/ppt-master/scripts/image_gen.py
+++ b/skills/ppt-master/scripts/image_gen.py
@@ -91,7 +91,7 @@ BACKEND_REGISTRY = {
"module": "backend_openai",
"tier": "core",
"label": "OpenAI / OpenAI-compatible",
- "default_model": "gpt-image-1",
+ "default_model": "gpt-image-2",
"key_hint": "OPENAI_API_KEY",
"aliases": ["openai-compatible", "openai_compatible"],
},