feat: add new models, mark deprecated ones

Add gpt-image-1.5, gpt-image-1-mini, gemini-2.5-flash-image,
gemini-3-pro-image-preview, and Imagen 4 variants to model registry.
Mark dall-e-2, dall-e-3, and gemini-2.0-flash as deprecated with
shutdown dates. Widen quality param to str, add 2:3/3:2 aspect ratios,
update default fallbacks from deprecated to current models.

Author: danielcopper

CONFIGURATION.md

@@ -53,7 +53,7 @@ Complete configuration reference for the Image Generation API.
 
 **`DEFAULT_MODEL`**
 - Default model for generation/editing when not specified in request
-- Example: `gpt-image-1`, `dall-e-3`
+- Example: `gpt-image-1`, `gpt-image-1.5`
 - Optional - if not set, first available model is used
 
 ### Response Configuration
@@ -109,27 +109,51 @@ Complete configuration reference for the Image Generation API.
 
 ### OpenAI Models
 
-**dall-e-2**
-- Aspect Ratios: `1:1` only
+**gpt-image-1.5** (latest)
+- Aspect Ratios: `1:1`, `2:3`, `3:2`
+- Quality: `auto`, `high`, `medium`, `low`
+- Max Images (n): 4
+- Editing: Yes (mask-based)
+
+**gpt-image-1**
+- Aspect Ratios: `1:1`, `16:9`, `9:16`
 - Quality: Not supported
 - Max Images (n): 4
 - Editing: Yes (mask-based)
 
-**dall-e-3**
+**gpt-image-1-mini**
+- Aspect Ratios: `1:1`, `16:9`, `9:16`
+- Quality: `low`, `medium`, `high`
+- Max Images (n): 4
+- Editing: Yes (mask-based)
+
+**dall-e-3** *(deprecated — shutting down May 12, 2026)*
 - Aspect Ratios: `1:1`, `16:9`, `9:16`
 - Quality: `standard`, `hd`
 - Max Images (n): 1
 - Editing: No
 
-**gpt-image-1**
-- Aspect Ratios: `1:1`, `16:9`, `9:16`
+**dall-e-2** *(deprecated — shutting down May 12, 2026)*
+- Aspect Ratios: `1:1` only
 - Quality: Not supported
 - Max Images (n): 4
 - Editing: Yes (mask-based)
 
 ### Google Gemini Models
 
-**gemini-2.0-flash-preview-image-generation**
+**gemini-2.5-flash-image** (recommended)
+- Aspect Ratios: `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `2:3`, `3:2`
+- Quality: Not supported
+- Max Images (n): 4
+- Editing: Yes (prompt-based)
+
+**gemini-3-pro-image-preview** (preview)
+- Aspect Ratios: `1:1`, `2:3`, `3:2`, `3:4`, `4:3`, `4:5`, `5:4`, `9:16`, `16:9`, `21:9`
+- Quality: Not supported
+- Max Images (n): 4
+- Editing: Yes (prompt-based)
+
+**gemini-2.0-flash-preview-image-generation** *(deprecated — shutting down March 31, 2026)*
 - Aspect Ratios: `1:1`, `16:9`, `9:16`, `4:3`, `3:4`
 - Quality: Not supported
 - Max Images (n): 4
@@ -187,7 +211,7 @@ Set `provider="openai"` or `provider="gemini"` in requests.
 
 ```json
 {
-  "model": "dall-e-3"  // or null for auto-select
+  "model": "gpt-image-1"  // or null for auto-select
 }
 ```
 
@@ -201,13 +225,15 @@ Leave `null` to use first available model for provider.
 }
 ```
 
-Options: `1:1`, `16:9` (landscape), `9:16` (portrait), `4:3`, `3:4`
+Options: `1:1`, `16:9` (landscape), `9:16` (portrait), `4:3`, `3:4`, `2:3` (tall portrait), `3:2` (wide landscape)
 
 Model compatibility:
-- dall-e-2: `1:1` only
-- dall-e-3: `1:1`, `16:9`, `9:16`
-- gpt-image-1: `1:1`, `16:9`, `9:16`
-- Gemini models: All ratios
+- gpt-image-1.5: `1:1`, `2:3`, `3:2`
+- gpt-image-1 / gpt-image-1-mini: `1:1`, `16:9`, `9:16`
+- gemini-2.5-flash-image: `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `2:3`, `3:2`
+- gemini-3-pro-image-preview: All 10 ratios including `4:5`, `5:4`, `21:9`
+- dall-e-3 (deprecated): `1:1`, `16:9`, `9:16`
+- dall-e-2 (deprecated): `1:1` only
 
 ### Quality
 
@@ -217,8 +243,11 @@ Model compatibility:
 }
 ```
 
-- `standard`: Faster generation
-- `hd`: Higher detail (dall-e-3 only)
+Values depend on model:
+- dall-e-3: `standard`, `hd`
+- gpt-image-1.5: `auto`, `high`, `medium`, `low`
+- gpt-image-1-mini: `low`, `medium`, `high`
+- Other models: quality parameter ignored
 
 ### Number of Images
 
@@ -230,6 +259,7 @@ Model compatibility:
 
 Constraints:
 - dall-e-3: max 1
+- imagen-4.0-ultra: max 1
 - Others: max 4
 
 ## Image Editing
@@ -238,7 +268,7 @@ The `/edit` endpoint supports two editing modes:
 
 ### Mask-based Editing (OpenAI)
 
-Used with `dall-e-2` and `gpt-image-1`. Requires:
+Used with `gpt-image-1`, `gpt-image-1.5`, `gpt-image-1-mini`, and `dall-e-2`. Requires:
 - `image`: Source image (upload or URL)
 - `mask`: PNG with transparent areas marking regions to edit
 - `prompt`: Description of what to generate in masked area
@@ -247,7 +277,7 @@ The mask should have transparent (alpha=0) pixels where editing should occur.
 
 ### Prompt-based Editing (Gemini)
 
-Used with `gemini-2.0-flash-preview-image-generation`. Requires:
+Used with `gemini-2.5-flash-image`, `gemini-3-pro-image-preview`, and other Gemini models. Requires:
 - `image`: Source image (upload or URL)
 - `prompt`: Natural language description of the edit
 

README.md

@@ -6,7 +6,7 @@ OpenAPI server for AI image generation and editing. Supports LiteLLM proxy (for
 
 - **Image Generation**: Create images from text prompts
 - **Image Editing**: Edit existing images with mask-based inpainting (OpenAI) or prompt-based editing (Gemini)
-- **Multi-Provider**: OpenAI (DALL-E 3, GPT-Image-1) and Google Gemini
+- **Multi-Provider**: OpenAI (GPT-Image-1.5, GPT-Image-1) and Google Gemini
 - **LiteLLM Integration**: Unified API with cost tracking
 - **Open WebUI Integration**: Auto-upload images to Open WebUI's file storage
 - **Dynamic Models**: Auto-discovers available models from LiteLLM
@@ -67,7 +67,7 @@ curl -X POST "http://localhost:8000/generate" \
   -H "Content-Type: application/json" \
   -d '{
     "prompt": "A mountain landscape at sunset",
-    "model": "dall-e-3",
+    "model": "gpt-image-1",
     "aspect_ratio": "16:9"
   }'
 ```
@@ -145,15 +145,19 @@ Currently applies to:
 
 | Model | Generation | Editing | Notes |
 |-------|------------|---------|-------|
-| dall-e-3 | Yes | No | HD quality, multiple aspect ratios |
+| gpt-image-1.5 | Yes | Yes (mask) | Latest, quality: auto/high/medium/low |
 | gpt-image-1 | Yes | Yes (mask) | Fast, supports inpainting |
-| dall-e-2 | Yes | Yes (mask) | Square only |
+| gpt-image-1-mini | Yes | Yes (mask) | Cost-efficient variant |
+| ~~dall-e-3~~ | Yes | No | **Deprecated** — shutting down May 12, 2026 |
+| ~~dall-e-2~~ | Yes | Yes (mask) | **Deprecated** — shutting down May 12, 2026 |
 
 ### Google Gemini
 
 | Model | Generation | Editing | Notes |
 |-------|------------|---------|-------|
-| gemini-2.0-flash-preview-image-generation | Yes | Yes (prompt) | Fast, prompt-based editing |
+| gemini-2.5-flash-image | Yes | Yes (prompt) | Fast, wide aspect ratio support |
+| gemini-3-pro-image-preview | Yes | Yes (prompt) | Preview, up to 4K resolution |
+| ~~gemini-2.0-flash-preview-image-generation~~ | Yes | Yes (prompt) | **Deprecated** — shutting down March 31, 2026 |
 
 See [CONFIGURATION.md](CONFIGURATION.md) for details.
 

TODO.md

@@ -0,0 +1,36 @@
+# TODO - openapi-image-gen
+
+## 1. Investigation & General Improvements
+Status: DONE
+
+Fix security and code quality issues found during review:
+- Use `secrets.compare_digest()` for bearer token comparison
+- Add basic SSRF guard to `storage_service.get_image()` (block metadata IPs, private ranges)
+- Fix CORS: don't combine `allow_origins=["*"]` with `allow_credentials=True`
+- Sanitize filenames extracted from URLs (even though UUIDs, harden the path)
+- Extract duplicated MIME type dict and OPENWEBUI_MODE response logic into helpers
+- Replace sync `open()` with `aiofiles` in async route handlers
+- Fix Dockerfile: add `COPY pyproject.toml .` so version reading works in container
+- Add async lock to model registry cache
+- Bump stale dependency versions where appropriate
+
+## 2. Add New Models
+Status: DONE
+
+Add models released since the last update:
+- OpenAI: gpt-image-1 variants, any new DALL-E models
+- Google: gemini-2.0-flash-preview-image-generation updates, newer Gemini image models, Imagen 3 updates
+- Update KNOWN_CAPABILITIES in model_registry.py
+- Update documentation tables
+- Test with LiteLLM model discovery
+
+## 3. Improve OWUI Image Display
+Status: PENDING
+
+Current approach returns HTMLResponse with base64-encoded `<img>` tag — a workaround
+that went through several iterations. OWUI has had updates since; investigate:
+- Check current OWUI tool response handling (artifact display, image rendering)
+- See if OWUI now natively supports image URLs or base64 in tool responses without HTML hack
+- Consider using OWUI's file upload API to store images there directly
+- Reduce unnecessary disk I/O (currently writes to disk then reads back for base64)
+- Test with latest OWUI version and document the recommended setup

app/api/routes/edit.py

@@ -268,6 +268,6 @@ def _get_default_edit_model(provider: str) -> str:
 
     # Fallback defaults
     if provider == "gemini":
-        return "gemini-2.0-flash-preview-image-generation"
+        return "gemini-2.5-flash-image"
     else:
         return "gpt-image-1"

app/api/routes/generate.py

@@ -314,11 +314,11 @@ def _get_default_model(provider: str) -> str:
     if not provider_models:
         # Fallback defaults
         if provider == "openai" or provider == "litellm":
-            return "dall-e-3"
+            return "gpt-image-1"
         elif provider == "gemini":
-            return "gemini-2.0-flash-preview-image-generation"
+            return "gemini-2.5-flash-image"
         else:
-            return "dall-e-3"
+            return "gpt-image-1"
 
     # Return first available model
     return provider_models[0].id

app/schemas/requests.py

@@ -34,17 +34,24 @@ class ImageRequest(BaseModel):
         examples=["openai/dall-e-3", "openai/gpt-image-1", "gemini/gemini-2.5-flash-image"],
     )
 
-    aspect_ratio: Literal["1:1", "16:9", "9:16", "4:3", "3:4"] = Field(
+    aspect_ratio: Literal["1:1", "16:9", "9:16", "4:3", "3:4", "2:3", "3:2"] = Field(
         default="1:1",
         description=(
             "Image aspect ratio. "
             "1:1 = square, 16:9 = landscape, 9:16 = portrait, "
-            "4:3 = classic landscape, 3:4 = classic portrait"
+            "4:3 = classic landscape, 3:4 = classic portrait, "
+            "2:3 = tall portrait, 3:2 = wide landscape"
         ),
     )
 
-    quality: Literal["standard", "hd"] = Field(
-        default="standard", description="Image quality. 'hd' available for dall-e-3 only"
+    quality: str = Field(
+        default="standard",
+        description=(
+            "Image quality. Values depend on model: "
+            "'standard'/'hd' (dall-e-3), "
+            "'auto'/'high'/'medium'/'low' (gpt-image-1.5). "
+            "Ignored by models that don't support quality."
+        ),
     )
 
     n: int = Field(

app/schemas/responses.py

@@ -45,6 +45,11 @@ class ModelCapabilities(BaseModel):
         default=False, description="Whether model supports quality parameter"
     )
 
+    quality_values: list[str] | None = Field(
+        default=None,
+        description="Accepted quality values (e.g., ['standard', 'hd'] or ['auto', 'high', 'medium', 'low'])",
+    )
+
     supports_aspect_ratios: list[str] = Field(
         default_factory=lambda: ["1:1"], description="Supported aspect ratios"
     )
@@ -60,6 +65,10 @@ class ModelCapabilities(BaseModel):
         description="Type of editing: 'mask' (OpenAI inpainting) or 'prompt' (Gemini natural language)",
     )
 
+    deprecated: str | None = Field(
+        default=None, description="Deprecation notice with shutdown date, or None if active"
+    )
+
 
 class ModelInfo(BaseModel):
     """

app/services/gemini_service.py

@@ -20,6 +20,8 @@ class GeminiService:
         "9:16": "9:16",
         "4:3": "4:3",
         "3:4": "3:4",
+        "2:3": "2:3",
+        "3:2": "3:2",
     }
 
     def __init__(self):

app/services/litellm_service.py

@@ -25,6 +25,8 @@ class LiteLLMService:
         "9:16": "1024x1792",
         "4:3": "1792x1024",
         "3:4": "1024x1792",
+        "2:3": "1024x1536",
+        "3:2": "1536x1024",
     }
 
     def __init__(self):