niahere 0.2.89 → 0.2.90

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "niahere",
-  "version": "0.2.89",
+  "version": "0.2.90",
   "description": "A personal AI assistant daemon — chat, scheduled jobs, persona system, extensible via skills.",
   "type": "module",
   "scripts": {

package/skills/image-generation/SKILL.md

@@ -14,12 +14,14 @@ General-purpose image generation skill supporting **OpenAI** (default) and **Gem
 ## Setup
 
 API keys in `~/.niahere/config.yaml`:
+
 ```yaml
 openai_api_key: sk-...
 gemini_api_key: AIza...
 ```
 
 Or set via CLI or environment variables:
+
 ```bash
 nia config set openai_api_key sk-...
 nia config set gemini_api_key AIza...
@@ -29,25 +31,39 @@ Keys are resolved in order: `--api-key` flag > env var (`$OPENAI_API_KEY` / `$GE
 
 ## Providers & Models
 
-| Provider | Default Model | Alternatives |
-|----------|--------------|--------------|
-| **OpenAI** (default) | `gpt-image-1.5` | `gpt-image-1`, `gpt-image-1-mini` |
-| **Gemini** | `gemini-3.1-flash-image-preview` | `gemini-3-pro-image-preview`, `gemini-2.5-flash-image` |
+| Provider             | Default Model                                    | Alternatives                                                                  |
+| -------------------- | ------------------------------------------------ | ----------------------------------------------------------------------------- |
+| **OpenAI** (default) | `gpt-image-2`                                    | `gpt-image-1.5`, `gpt-image-1-mini`                                           |
+| **Gemini**           | `gemini-3.1-flash-image-preview` (Nano Banana 2) | `gemini-3-pro-image-preview` (Nano Banana Pro), `gemini-2.5-flash-image` (GA) |
+
+`gpt-image-2` (Apr 2026) is the current OpenAI flagship — native reasoning, up to 16 reference images, native 2K. `dall-e-2`/`dall-e-3` were shut off on the API on May 12, 2026.
 
-Note: `dall-e-2` and `dall-e-3` are deprecated (EOL May 2026). Use `gpt-image-1.5` instead.
+## Per-image pricing (May 2026)
+
+| Model                            | Price                                                                                |
+| -------------------------------- | ------------------------------------------------------------------------------------ |
+| `gpt-image-2`                    | ~$0.006 (low) → $0.053 (medium 1024) → $0.35 (high 4K). Token-based; Batch API −50%. |
+| `gpt-image-1.5`                  | ~20% cheaper than gpt-image-1 across tiers                                           |
+| `gpt-image-1`                    | low $0.011, medium $0.042, high $0.167 (1024²)                                       |
+| `gemini-3-pro-image-preview`     | $0.134 (1K/2K), $0.24 (4K)                                                           |
+| `gemini-3.1-flash-image-preview` | $0.045 (512), $0.067 (1K), $0.101 (2K), $0.151 (4K)                                  |
+| `gemini-2.5-flash-image`         | $0.039 standard, $0.0195 batch                                                       |
 
 ## Quick Reference
 
 ```bash
-SCRIPT="/Users/aman/.shared/skills/image-generation/scripts/generate_image.py"
+SCRIPT="$(dirname "$0")/scripts/generate_image.py"   # or the absolute path under your skills dir
 
-# Basic generation (OpenAI, default)
+# Basic generation (OpenAI gpt-image-2, default)
 python3 $SCRIPT --prompt "A sunset over mountains"
 
 # High quality
 python3 $SCRIPT --prompt "Oil painting of a forest" --quality high
 
-# With aspect ratio
+# 2K output (OpenAI gpt-image-2; Gemini 2K via --resolution)
+python3 $SCRIPT --prompt "Detailed cityscape" --resolution 2K
+
+# Aspect ratio
 python3 $SCRIPT --prompt "Portrait photo" --aspect-ratio 3:4
 
 # With reference image (OpenAI edit mode)
@@ -56,6 +72,10 @@ python3 $SCRIPT --prompt "Add a rainbow to this scene" --reference photo.png
 # Gemini provider
 python3 $SCRIPT --provider gemini --prompt "Watercolor sunset" --aspect-ratio 16:9
 
+# Gemini 4K (use Pro — 3.1 Flash currently ignores 2K/4K and returns ~1K)
+python3 $SCRIPT --provider gemini --model gemini-3-pro-image-preview \
+  --prompt "Cinematic landscape" --resolution 4K --aspect-ratio 16:9
+
 # Gemini with reference
 python3 $SCRIPT --provider gemini --reference face.png \
   --prompt "Same person sitting in a cafe, natural lighting" --aspect-ratio 9:16
@@ -66,17 +86,24 @@ python3 $SCRIPT --prompt "A cat" --output /path/to/output/
 
 ## Aspect Ratios
 
-| Use Case | Ratio | Notes |
-|----------|-------|-------|
-| Square / social | `1:1` | Default |
-| Portrait | `3:4` or `2:3` | Vertical |
-| Landscape | `4:3` or `16:9` | Wide |
-| Phone / story | `9:16` | Vertical tall |
-| Ultrawide | `21:9` | Cinematic |
+| Use Case        | Ratio           | Notes         |
+| --------------- | --------------- | ------------- |
+| Square / social | `1:1`           | Default       |
+| Portrait        | `3:4` or `2:3`  | Vertical      |
+| Landscape       | `4:3` or `16:9` | Wide          |
+| Phone / story   | `9:16`          | Vertical tall |
+| Ultrawide       | `21:9`          | Cinematic     |
+
+OpenAI maps ratios to closest supported size. At `--resolution 1K` (default): `1024x1024`, `1024x1536`, `1536x1024`. At `--resolution 2K`: `2048x2048`, `1536x2048`, `2048x1536` (gpt-image-2 only; dims must be multiples of 16, max ~2048).
+
+## Resolution (`--resolution`)
 
-OpenAI maps ratios to closest supported size (`1024x1024`, `1024x1536`, `1536x1024`).
+- OpenAI: `1K` (default) or `2K` (gpt-image-2 only).
+- Gemini: `1K` (default), `2K`, or `4K`. **Caveat:** `gemini-3.1-flash-image-preview` currently ignores 2K/4K and returns ~1K — use `gemini-3-pro-image-preview` for true 2K/4K.
 
-## OpenAI Quality (gpt-image-1 only)
+## OpenAI Quality (`--quality`)
+
+Applies to all `gpt-image-*` models.
 
 - `auto` (default) — let the model decide
 - `high` — best quality, slower
@@ -100,6 +127,7 @@ For photorealistic results, use structured JSON prompts covering separate concer
 ```
 
 Key principles:
+
 1. **Separate concerns** — one aspect per block
 2. **Specify camera** — lens mm and aperture drive realism
 3. **Light direction** — "soft light from upper right" > "good lighting"
@@ -108,13 +136,14 @@ Key principles:
 
 ## Provider Selection Guide
 
-| Need | Use |
-|------|-----|
-| General image gen, highest quality | OpenAI `gpt-image-1.5` |
-| Budget-friendly | OpenAI `gpt-image-1-mini` |
-| Reference-based identity (same face) | Gemini (better at preserving identity from reference) |
-| Image editing / inpainting | OpenAI edit mode (`--reference`) |
-| Free tier / no OpenAI key | Gemini |
+| Need                                 | Use                                                          |
+| ------------------------------------ | ------------------------------------------------------------ |
+| General image gen, highest quality   | OpenAI `gpt-image-2`                                         |
+| Budget-friendly                      | OpenAI `gpt-image-1-mini` or Gemini `gemini-2.5-flash-image` |
+| Reference-based identity (same face) | Gemini (better at preserving identity from reference)        |
+| Image editing / inpainting           | OpenAI edit mode (`--reference`)                             |
+| 4K output                            | Gemini `gemini-3-pro-image-preview` (`--resolution 4K`)      |
+| Free tier / no OpenAI key            | Gemini                                                       |
 
 ## Combining with Bella
 

package/skills/image-generation/scripts/generate_image.py

@@ -3,21 +3,25 @@
 General-purpose image generation using OpenAI (default) or Gemini.
 
 Supports:
-  - OpenAI: gpt-image-1.5 (default), gpt-image-1, gpt-image-1-mini
+  - OpenAI: gpt-image-2 (default), gpt-image-1.5, gpt-image-1-mini
   - Gemini: gemini-3.1-flash-image-preview (default), gemini-3-pro-image-preview, gemini-2.5-flash-image
 
 Usage:
   # OpenAI (default)
   python3 generate_image.py --prompt "A sunset over mountains"
 
+  # OpenAI 2K output (gpt-image-2 only)
+  python3 generate_image.py --prompt "Detailed cityscape" --resolution 2K
+
   # OpenAI with reference image (edit mode)
   python3 generate_image.py --prompt "Add a hot air balloon" --reference photo.png
 
   # Gemini
   python3 generate_image.py --provider gemini --prompt "A sunset over mountains"
 
-  # Gemini with reference image
-  python3 generate_image.py --provider gemini --prompt "Same person in a cafe" --reference face.png
+  # Gemini Pro at 4K
+  python3 generate_image.py --provider gemini --model gemini-3-pro-image-preview \\
+    --prompt "Cinematic landscape" --resolution 4K --aspect-ratio 16:9
 """
 
 from __future__ import annotations
@@ -40,17 +44,26 @@ NIA_CONFIG = NIA_HOME / "config.yaml"
 TIMESTAMP_FORMAT = "%Y%m%d_%H%M%S"
 
 # --- Provider defaults ---
-OPENAI_DEFAULT_MODEL = "gpt-image-1.5"
+OPENAI_DEFAULT_MODEL = "gpt-image-2"
 GEMINI_DEFAULT_MODEL = "gemini-3.1-flash-image-preview"
 
 DEFAULT_ASPECT_RATIO = "1:1"
 ALLOWED_ASPECT_RATIOS = (
-    "1:1", "3:4", "4:3", "9:16", "16:9",
-    "2:3", "3:2", "4:5", "5:4", "21:9",
+    "1:1",
+    "3:4",
+    "4:3",
+    "9:16",
+    "16:9",
+    "2:3",
+    "3:2",
+    "4:5",
+    "5:4",
+    "21:9",
 )
 
-# OpenAI size mappings (closest match for aspect ratio)
-OPENAI_SIZE_MAP = {
+# OpenAI size mappings per resolution (closest match for aspect ratio).
+# 2K caps at 2048; dims must be multiples of 16. gpt-image-2 only.
+OPENAI_SIZE_MAP_1K = {
     "1:1": "1024x1024",
     "3:4": "1024x1536",
     "4:3": "1536x1024",
@@ -62,10 +75,25 @@ OPENAI_SIZE_MAP = {
     "5:4": "1536x1024",
     "21:9": "1536x1024",
 }
+OPENAI_SIZE_MAP_2K = {
+    "1:1": "2048x2048",
+    "3:4": "1536x2048",
+    "4:3": "2048x1536",
+    "9:16": "1536x2048",
+    "16:9": "2048x1536",
+    "2:3": "1536x2048",
+    "3:2": "2048x1536",
+    "4:5": "1536x2048",
+    "5:4": "2048x1536",
+    "21:9": "2048x1536",
+}
 
-# OpenAI quality options
+# OpenAI quality options (applies to all gpt-image-* models)
 OPENAI_QUALITIES = ("auto", "high", "medium", "low")
 
+# Resolution choices. OpenAI: 1K|2K (2K requires gpt-image-2). Gemini: 1K|2K|4K.
+ALLOWED_RESOLUTIONS = ("1K", "2K", "4K")
+
 
 def safe_mime(path: str) -> str:
     mime, _ = mimetypes.guess_type(path)
@@ -92,6 +120,7 @@ def read_config_key(key: str) -> str:
         return ""
     try:
         import importlib
+
         yaml = importlib.import_module("yaml")
         with NIA_CONFIG.open("r") as f:
             config = yaml.safe_load(f)
@@ -110,10 +139,7 @@ def resolve_api_key(provider: str, cli_key: str | None) -> str:
         return cli_key
 
     if provider == "openai":
-        return (
-            os.environ.get("OPENAI_API_KEY", "")
-            or read_config_key("openai_api_key")
-        )
+        return os.environ.get("OPENAI_API_KEY", "") or read_config_key("openai_api_key")
     else:
         return (
             os.environ.get("GEMINI_API_KEY", "")
@@ -124,6 +150,7 @@ def resolve_api_key(provider: str, cli_key: str | None) -> str:
 
 # --- OpenAI Generation ---
 
+
 def generate_openai(
     api_key: str,
     prompt: str,
@@ -139,18 +166,15 @@ def generate_openai(
     return _openai_generate(api_key, prompt, model, size, quality, n)
 
 
-def _openai_generate(
-    api_key: str, prompt: str, model: str, size: str, quality: str, n: int
-) -> tuple[bytes, str]:
+def _openai_generate(api_key: str, prompt: str, model: str, size: str, quality: str, n: int) -> tuple[bytes, str]:
     url = "https://api.openai.com/v1/images/generations"
     payload: dict = {
         "model": model,
         "prompt": prompt,
         "n": n,
         "size": size,
-        "response_format": "b64_json",
     }
-    if model == "gpt-image-1":
+    if model.startswith("gpt-image-"):
         payload["quality"] = quality
 
     req = urllib.request.Request(
@@ -166,8 +190,7 @@ def _openai_generate(
 
 
 def _openai_edit(
-    api_key: str, prompt: str, reference_path: str, model: str, size: str,
-    quality: str, n: int
+    api_key: str, prompt: str, reference_path: str, model: str, size: str, quality: str, n: int
 ) -> tuple[bytes, str]:
     """Use OpenAI images/edits endpoint with a reference image."""
     import io
@@ -184,9 +207,7 @@ def _openai_edit(
         filename = Path(filepath).name
         mime = safe_mime(filepath)
         body.write(f"--{boundary}\r\n".encode())
-        body.write(
-            f'Content-Disposition: form-data; name="{name}"; filename="{filename}"\r\n'.encode()
-        )
+        body.write(f'Content-Disposition: form-data; name="{name}"; filename="{filename}"\r\n'.encode())
         body.write(f"Content-Type: {mime}\r\n\r\n".encode())
         with open(filepath, "rb") as f:
             body.write(f.read())
@@ -197,7 +218,7 @@ def _openai_edit(
     add_field("model", model)
     add_field("n", str(n))
     add_field("size", size)
-    if model == "gpt-image-1":
+    if model.startswith("gpt-image-"):
         add_field("quality", quality)
     body.write(f"--{boundary}--\r\n".encode())
 
@@ -220,9 +241,7 @@ def _openai_request(req: urllib.request.Request) -> tuple[bytes, str]:
             response = json.loads(resp.read().decode("utf-8"))
     except urllib.error.HTTPError as e:
         detail = e.read().decode("utf-8", errors="ignore")
-        raise RuntimeError(
-            f"OpenAI API error (HTTP {e.code}): {detail or e.reason}"
-        ) from e
+        raise RuntimeError(f"OpenAI API error (HTTP {e.code}): {detail or e.reason}") from e
 
     data_list = response.get("data", [])
     if not data_list:
@@ -237,33 +256,40 @@ def _openai_request(req: urllib.request.Request) -> tuple[bytes, str]:
 
 # --- Gemini Generation ---
 
+
 def generate_gemini(
     api_key: str,
     prompt: str,
     model: str,
     aspect_ratio: str,
+    resolution: str = "1K",
     reference_path: str | None = None,
 ) -> tuple[bytes, str]:
     """Generate image via Gemini API."""
-    url = (
-        "https://generativelanguage.googleapis.com/"
-        f"v1beta/models/{model}:generateContent?key={api_key}"
-    )
+    url = f"https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent?key={api_key}"
 
     parts: list[dict] = []
     if reference_path and Path(reference_path).is_file():
-        parts.append({
-            "inlineData": {
-                "mimeType": safe_mime(reference_path),
-                "data": encode_file(reference_path),
+        parts.append(
+            {
+                "inlineData": {
+                    "mimeType": safe_mime(reference_path),
+                    "data": encode_file(reference_path),
+                }
             }
-        })
+        )
     parts.append({"text": prompt})
 
+    image_config: dict = {"aspectRatio": aspect_ratio}
+    if resolution and resolution != "1K":
+        # gemini-3.1-flash-image-preview currently ignores this and returns ~1K (Google bug);
+        # gemini-3-pro-image-preview honors it. Send anyway so it works once Flash is fixed.
+        image_config["imageSize"] = resolution
+
     payload = {
         "contents": [{"parts": parts}],
         "generationConfig": {
-            "imageConfig": {"aspectRatio": aspect_ratio},
+            "imageConfig": image_config,
             "responseModalities": ["TEXT", "IMAGE"],
         },
     }
@@ -280,9 +306,7 @@ def generate_gemini(
             response = json.loads(resp.read().decode("utf-8"))
     except urllib.error.HTTPError as e:
         detail = e.read().decode("utf-8", errors="ignore")
-        raise RuntimeError(
-            f"Gemini API error (HTTP {e.code}): {detail or e.reason}"
-        ) from e
+        raise RuntimeError(f"Gemini API error (HTTP {e.code}): {detail or e.reason}") from e
 
     candidates = response.get("candidates", [])
     if not candidates:
@@ -298,13 +322,12 @@ def generate_gemini(
         if mime and img_data:
             return base64.b64decode(img_data), mime
 
-    raise RuntimeError(
-        f"No image in Gemini response: {json.dumps(response, indent=2)}"
-    )
+    raise RuntimeError(f"No image in Gemini response: {json.dumps(response, indent=2)}")
 
 
 # --- CLI ---
 
+
 def main() -> None:
     parser = argparse.ArgumentParser(
         description="Generate images using OpenAI (default) or Gemini.",
@@ -319,39 +342,58 @@ Examples:
         """,
     )
     parser.add_argument(
-        "--provider", choices=["openai", "gemini"], default="openai",
+        "--provider",
+        choices=["openai", "gemini"],
+        default="openai",
         help="Image generation provider. Default: openai.",
     )
     parser.add_argument(
-        "--prompt", required=True,
+        "--prompt",
+        required=True,
         help="Text prompt describing the image to generate.",
     )
     parser.add_argument(
-        "--reference", default=None,
+        "--reference",
+        default=None,
         help="Path to a reference image. OpenAI uses edit mode; Gemini includes it as context.",
     )
     parser.add_argument(
-        "--model", default=None,
+        "--model",
+        default=None,
         help=f"Model override. Defaults: OpenAI={OPENAI_DEFAULT_MODEL}, Gemini={GEMINI_DEFAULT_MODEL}.",
     )
     parser.add_argument(
-        "--aspect-ratio", default=DEFAULT_ASPECT_RATIO, choices=ALLOWED_ASPECT_RATIOS,
+        "--aspect-ratio",
+        default=DEFAULT_ASPECT_RATIO,
+        choices=ALLOWED_ASPECT_RATIOS,
         help=f"Aspect ratio. Default: {DEFAULT_ASPECT_RATIO}.",
     )
     parser.add_argument(
-        "--quality", default="auto", choices=OPENAI_QUALITIES,
-        help="OpenAI quality (gpt-image-1 only). Default: auto.",
+        "--quality",
+        default="auto",
+        choices=OPENAI_QUALITIES,
+        help="OpenAI quality (all gpt-image-* models). Default: auto.",
+    )
+    parser.add_argument(
+        "--resolution",
+        default="1K",
+        choices=ALLOWED_RESOLUTIONS,
+        help="Output resolution. OpenAI: 1K|2K (2K needs gpt-image-2). Gemini: 1K|2K|4K (2K/4K reliable on Pro only). Default: 1K.",
     )
     parser.add_argument(
-        "--n", type=int, default=1,
+        "--n",
+        type=int,
+        default=1,
         help="Number of images (OpenAI only). Default: 1.",
     )
     parser.add_argument(
-        "--output", default=None,
+        "--output",
+        default=None,
         help="Output path. Directory = timestamped file. Default: /tmp/.",
     )
     parser.add_argument(
-        "--api-key", default=None,
+        "--api-key",
+        default=None,
         help="API key override. Otherwise reads from env var or ~/.niahere/config.yaml.",
     )
     args = parser.parse_args()
@@ -364,8 +406,7 @@ Examples:
         config_key = "openai_api_key" if provider == "openai" else "gemini_api_key"
         env_var = "OPENAI_API_KEY" if provider == "openai" else "GEMINI_API_KEY"
         raise SystemExit(
-            f"Missing API key. Provide --api-key, set {env_var} in environment, "
-            f"or add {config_key} to {NIA_CONFIG}."
+            f"Missing API key. Provide --api-key, set {env_var} in environment, or add {config_key} to {NIA_CONFIG}."
         )
 
     if args.reference and not Path(args.reference).expanduser().is_file():
@@ -375,15 +416,29 @@ Examples:
 
     try:
         if provider == "openai":
-            size = OPENAI_SIZE_MAP.get(args.aspect_ratio, "1024x1024")
+            if args.resolution == "4K":
+                raise SystemExit("OpenAI does not support 4K via this script — use --provider gemini --resolution 4K.")
+            if args.resolution == "2K" and model != "gpt-image-2":
+                raise SystemExit(f"2K is only supported on gpt-image-2 (got --model {model}).")
+            size_map = OPENAI_SIZE_MAP_2K if args.resolution == "2K" else OPENAI_SIZE_MAP_1K
+            size = size_map.get(args.aspect_ratio, size_map["1:1"])
             image_data, mime = generate_openai(
-                api_key=api_key, prompt=args.prompt, model=model,
-                size=size, quality=args.quality, reference_path=ref, n=args.n,
+                api_key=api_key,
+                prompt=args.prompt,
+                model=model,
+                size=size,
+                quality=args.quality,
+                reference_path=ref,
+                n=args.n,
             )
         else:
             image_data, mime = generate_gemini(
-                api_key=api_key, prompt=args.prompt, model=model,
-                aspect_ratio=args.aspect_ratio, reference_path=ref,
+                api_key=api_key,
+                prompt=args.prompt,
+                model=model,
+                aspect_ratio=args.aspect_ratio,
+                resolution=args.resolution,
+                reference_path=ref,
             )
 
         ext = ".png" if "png" in mime else ".jpg"
@@ -391,7 +446,7 @@ Examples:
         out.parent.mkdir(parents=True, exist_ok=True)
         out.write_bytes(image_data)
         print(f"Saved: {out}")
-        print(f"Provider: {provider} | Model: {model} | Size/Ratio: {args.aspect_ratio}")
+        print(f"Provider: {provider} | Model: {model} | Ratio: {args.aspect_ratio} | Resolution: {args.resolution}")
     except Exception as exc:
         print(f"Error: {exc}", file=sys.stderr)
         raise SystemExit(1) from exc