egaki 0.2.0 → 0.4.0

package/CHANGELOG.md

@@ -2,6 +2,76 @@
 
 All notable changes to this project will be documented in this file.
 
+## 0.4.0
+
+1. **ChatGPT login for OpenAI image generation** — use your ChatGPT subscription in `egaki` without a separate OpenAI platform API key:
+
+   ```bash
+   egaki login --provider chatgpt
+   egaki image "a dreamy studio ghibli style bakery at sunrise" -m gpt-image-1.5 -o bakery.png
+   ```
+
+   OpenAI image requests authenticated this way now follow the same Codex backend flow used by ChatGPT instead of the normal Image API path.
+
+2. **ChatGPT-backed image editing** — edit existing images with OpenAI image models through the ChatGPT/Codex backend:
+
+   ```bash
+   egaki image "change the red jacket to a blue jacket" -m gpt-image-1.5 --input portrait.png -o portrait-blue.png
+   egaki image "turn this product shot into a clay render" -m gpt-image-1.5 --input product.png -o product-clay.png
+   ```
+
+   Multiple input images are supported. For this backend path, `egaki` now explicitly rejects unsupported options like `--seed`, `--mask`, and multi-image output instead of pretending they work.
+
+3. **Aspect ratio support for ChatGPT image generation** — supported ChatGPT/OpenAI aspect ratios now map to the backend's real size controls:
+
+   ```bash
+   egaki image "wide landscape matte painting" -m gpt-image-1.5 --aspect-ratio 3:2 -o wide.png
+   egaki image "book cover concept art" -m gpt-image-1.5 --aspect-ratio 2:3 -o cover.png
+   ```
+
+   Supported ratios on this path are `1:1`, `3:2`, and `2:3`.
+
+4. **`egaki models` now shows login availability** — model discovery output includes whether each provider is currently usable from your configured credentials:
+
+   ```bash
+   egaki models
+   egaki models --json | jq '.[] | {id, provider, auth}'
+   ```
+
+   The output now includes `auth.available` and `auth.source` (`env`, `stored`, `oauth`, or `none`) so you can see which model families are ready to use.
+
+## 0.3.0
+
+1. **Google Vertex AI provider** — use Google Cloud billing instead of AI Studio by
+   prefixing model IDs with `vertex/`:
+
+   ```bash
+   egaki login --provider vertex --key AIza...
+   egaki image "product shot on marble" -m vertex/imagen-4.0-generate-001 -o product.png
+   egaki image "editorial portrait" -m vertex/gemini-3.1-flash-image-preview --aspect-ratio 4:5
+   egaki video "storm over mountains" -m vertex/veo-3.1-fast-generate-001 --duration 6 -o storm.mp4
+   ```
+
+   Bare model IDs (e.g. `imagen-4.0-generate-001`) continue to route through Google AI
+   Studio as before. `vertex/` prefix routes through `@ai-sdk/google-vertex` using your
+   `GOOGLE_VERTEX_API_KEY`. The two providers are fully independent — having one key does
+   not affect the other.
+
+   Supported Vertex models:
+   - `vertex/imagen-4.0-generate-001`, `vertex/imagen-4.0-ultra-generate-001`, `vertex/imagen-4.0-fast-generate-001`
+   - `vertex/gemini-2.5-flash-image`, `vertex/gemini-3-pro-image-preview`, `vertex/gemini-3.1-flash-image-preview`
+   - `vertex/veo-3.1-generate-001`, `vertex/veo-3.1-fast-generate-001`
+
+2. **`egaki models --provider vertex`** — list only Vertex models:
+
+   ```bash
+   egaki models --provider vertex
+   ```
+
+3. **Fixed confusing error for Vertex without a key** — if you attempt a `vertex/` model
+   without a `GOOGLE_VERTEX_API_KEY` configured, you now get a clear error pointing you
+   to set one up instead of a cryptic upstream failure.
+
 ## 0.2.0
 
 1. **New `egaki video` command** — generate videos from text prompts or still images.

package/README.md

@@ -20,6 +20,14 @@ egaki image "a watercolor fox reading a map" -o fox.png
 egaki video "a paper boat drifting on a calm lake at sunrise" -o boat.mp4
 ```
 
+### Quick start with ChatGPT auth
+
+```bash
+egaki login --provider chatgpt
+egaki image "a dreamy studio ghibli style bakery at sunrise" -m gpt-image-1.5 -o bakery.png
+egaki image "change the red jacket to a blue jacket" -m gpt-image-1.5 --input portrait.png -o portrait-blue.png
+```
+
 ## CLI examples
 
 ### Generate an image from text
@@ -34,6 +42,7 @@ egaki image "isometric floating city, detailed, soft colors" -m imagen-4.0-gener
 ```bash
 egaki image "add a red scarf and make it winter" --input portrait.jpg -o portrait-winter.png
 egaki image "turn this into a manga panel" --input https://example.com/photo.jpg -o manga.png
+egaki image "change the red square to a blue square" -m gpt-image-1.5 --input input.png -o output.png
 ```
 
 ### Inpainting with a mask
@@ -53,6 +62,15 @@ egaki image "minimal logo concepts for a cat cafe" -n 4 -o logo.png
 ```bash
 egaki image "cyberpunk alley at night" --aspect-ratio 16:9
 egaki image "polaroid-style travel photo" --aspect-ratio 4:5
+egaki image "wide landscape matte painting" -m gpt-image-1.5 --aspect-ratio 3:2 -o wide.png
+```
+
+### Use Google Cloud billing via Vertex AI
+
+```bash
+egaki login --provider vertex --key AIza...
+egaki image "editorial sneaker photo on white seamless" -m vertex/imagen-4.0-generate-001 -o sneaker.png
+egaki video "storm over mountains" -m vertex/veo-3.1-fast-generate-001 --duration 6 -o storm.mp4
 ```
 
 ### Pipe image output to other tools
@@ -67,7 +85,7 @@ egaki image "flat icon of a fox" --stdout | magick - -resize 512x512 fox-icon.pn
 
 ```bash
 egaki video "a paper boat drifting on a calm lake at sunrise" -o boat.mp4
-egaki video "timelapse of a stormy sea, cinematic" -m google/veo-3.1-generate-001 --duration 8 -o storm.mp4
+egaki video "timelapse of a stormy sea, cinematic" -m veo-3.1-generate-001 --duration 8 -o storm.mp4
 ```
 
 ### Generate with a cheap model
@@ -88,7 +106,7 @@ egaki video "slowly animate the clouds" --input photo.jpg -m klingai/kling-v2.6-
 
 ```bash
 egaki video "aerial drone shot over a city grid" \
-  -m google/veo-3.1-fast-generate-001 \
+  -m veo-3.1-fast-generate-001 \
   --aspect-ratio 16:9 \
   --resolution 1080p \
   --duration 6 \
@@ -175,8 +193,9 @@ egaki image "mascot variations, flat vector look" \
 
 | Model ID | Best for | Example command |
 | --- | --- | --- |
-| `google/veo-3.1-generate-001` | Highest quality video with audio, up to 4K | `egaki video "rainy Tokyo street at night" -m google/veo-3.1-generate-001 --duration 8 -o tokyo.mp4` |
-| `google/veo-3.1-fast-generate-001` | Fast Veo with 720p–4K, good for iteration | `egaki video "abstract paint patterns" -m google/veo-3.1-fast-generate-001 --duration 5 -o paint.mp4` |
+| `veo-3.1-generate-001` | Highest quality video with audio, up to 4K | `egaki video "rainy Tokyo street at night" -m veo-3.1-generate-001 --duration 8 -o tokyo.mp4` |
+| `veo-3.1-fast-generate-001` | Fast Veo with 720p–4K, good for iteration | `egaki video "abstract paint patterns" -m veo-3.1-fast-generate-001 --duration 5 -o paint.mp4` |
+| `vertex/veo-3.1-generate-001` | Same as above, routed through Vertex AI | `egaki video "rainy Tokyo street" -m vertex/veo-3.1-generate-001 --duration 8 -o tokyo.mp4` |
 | `klingai/kling-v2.5-turbo-t2v` | Cheap, fast Kling text-to-video | `egaki video "a paper boat on a pond" -m klingai/kling-v2.5-turbo-t2v --duration 5 -o boat.mp4` |
 | `bytedance/seedance-v1.5-pro` | Bytedance, audio support, three resolutions | `egaki video "timelapse of clouds above mountains" -m bytedance/seedance-v1.5-pro -o clouds.mp4` |
 | `xai/grok-imagine-video` | xAI video generation, cheap for short clips | `egaki video "a dog catching a frisbee" -m xai/grok-imagine-video --duration 3 -o dog.mp4` |
@@ -189,6 +208,7 @@ egaki image "mascot variations, flat vector look" \
 - **BFL image models (`flux-*`)**: Kontext/Pro variants via AI Gateway subscription
 - **Recraft models (`recraft-*`)**: v2/v3/v4 families available via AI Gateway subscription
 - **xAI image models (`grok-imagine-*`)**: Grok image generation via AI Gateway subscription
+- **Vertex models (`vertex/*`)**: same models as Google AI Studio, routed through Vertex AI / Google Cloud billing
 - **Google Veo video models**: up to 4K, audio optional, duration 4–8s
 - **Kling video models**: mode (std/pro), audio on v2.6+, image-to-video support
 - **Bytedance Seedance**: 480p–1080p, audio support on v1.5-pro
@@ -199,7 +219,9 @@ egaki image "mascot variations, flat vector look" \
 egaki supports **both** authentication modes:
 
 - **BYOK (bring your own keys):** add provider keys with `egaki login` per provider.
+- **ChatGPT auth:** log in with `egaki login --provider chatgpt` to use the ChatGPT/Codex backend for supported OpenAI image generation and editing flows.
 - **Egaki subscription:** use one `egaki_...` key to access all supported models without managing keys for each provider.
+- **Google vs Vertex:** bare model IDs (e.g. `imagen-4.0-generate-001`) use Google AI Studio. Prefix with `vertex/` (e.g. `vertex/imagen-4.0-generate-001`) to route through Vertex AI / Google Cloud billing.
 
 ```bash
 # Subscribe and get a checkout URL
@@ -211,8 +233,13 @@ egaki subscribe --plan pro
 # Save your Egaki key after checkout
 egaki login --provider egaki --key egaki_...
 
-# BYOK example (direct provider key)
+# BYOK examples (direct provider keys)
 egaki login --provider google --key AIza...
+egaki login --provider vertex --key AIza...
+
+# ChatGPT OAuth for Codex-backed image generation/editing
+egaki login --provider chatgpt
+egaki image "turn this product shot into a clay render" -m gpt-image-1.5 --input product.png -o product-clay.png
 
 # Check subscription usage / cancel
 egaki usage

package/dist/chatgpt-auth.d.ts

@@ -0,0 +1,25 @@
+export type ChatGptAuth = {
+    email?: string;
+    accountId?: string;
+    plan?: string;
+    refresh: string;
+    access: string;
+    expires: number;
+};
+/** Extract the ChatGPT plan type (e.g. "plus", "pro") from the access token. */
+export declare function extractPlanType(auth: ChatGptAuth): string | undefined;
+/**
+ * Run the full ChatGPT OAuth browser flow with clack UI.
+ * Returns the ChatGptAuth object to be stored in credentials.json.
+ */
+export declare function chatGptOAuthLogin(): Promise<ChatGptAuth>;
+/**
+ * Refresh an expired ChatGPT access token using the refresh token.
+ * Returns the updated auth object, or an Error if refresh fails.
+ */
+export declare function refreshChatGptToken(auth: ChatGptAuth): Promise<ChatGptAuth | Error>;
+/**
+ * Refresh the auth object if needed and return the full updated state.
+ */
+export declare function getValidChatGptAuth(auth: ChatGptAuth, onRefresh?: (updated: ChatGptAuth) => void): Promise<ChatGptAuth | Error>;
+//# sourceMappingURL=chatgpt-auth.d.ts.map

package/dist/chatgpt-auth.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"chatgpt-auth.d.ts","sourceRoot":"","sources":["../src/chatgpt-auth.ts"],"names":[],"mappings":"AAmBA,MAAM,MAAM,WAAW,GAAG;IACxB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;CAChB,CAAA;AAgGD,gFAAgF;AAChF,wBAAgB,eAAe,CAAC,IAAI,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAOrE;AA8JD;;;GAGG;AACH,wBAAsB,iBAAiB,IAAI,OAAO,CAAC,WAAW,CAAC,CA8C9D;AAED;;;GAGG;AACH,wBAAsB,mBAAmB,CAAC,IAAI,EAAE,WAAW,GAAG,OAAO,CAAC,WAAW,GAAG,KAAK,CAAC,CAyBzF;AAED;;GAEG;AACH,wBAAsB,mBAAmB,CACvC,IAAI,EAAE,WAAW,EACjB,SAAS,CAAC,EAAE,CAAC,OAAO,EAAE,WAAW,KAAK,IAAI,GACzC,OAAO,CAAC,WAAW,GAAG,KAAK,CAAC,CAS9B"}

package/dist/chatgpt-auth.js

@@ -3,8 +3,8 @@
 // Tokens are stored inside credentials.json under the 'chatgpt' key as a
 // structured object (not a plain string like other providers).
 //
-// The access token is a standard OpenAI bearer token that works with the
-// Image API, so we can inject it as OPENAI_API_KEY for the SDK to pick up.
+// Tokens are stored so egaki can reuse the ChatGPT login for Codex-style
+// backend requests and refresh them when needed.
 import { createServer } from 'node:http';
 import { spawn } from 'node:child_process';
 import { spinner, log, note } from '@clack/prompts';
@@ -55,6 +55,13 @@ function extractEmail(tokens) {
     return ((typeof idClaims?.email === 'string' ? idClaims.email : undefined) ??
         (typeof accessClaims?.email === 'string' ? accessClaims.email : undefined));
 }
+function extractPlanTypeFromTokens(tokens) {
+    const idClaims = tokens.id_token ? parseJwtClaims(tokens.id_token) : undefined;
+    const accessClaims = parseJwtClaims(tokens.access_token);
+    const idAuth = idClaims?.['https://api.openai.com/auth'];
+    const accessAuth = accessClaims?.['https://api.openai.com/auth'];
+    return idAuth?.chatgpt_plan_type ?? accessAuth?.chatgpt_plan_type;
+}
 function extractAccountId(tokens) {
     const idClaims = tokens.id_token ? parseJwtClaims(tokens.id_token) : undefined;
     const accessClaims = parseJwtClaims(tokens.access_token);
@@ -71,6 +78,8 @@ function extractAccountId(tokens) {
 }
 /** Extract the ChatGPT plan type (e.g. "plus", "pro") from the access token. */
 export function extractPlanType(auth) {
+    if (auth.plan)
+        return auth.plan;
     const claims = parseJwtClaims(auth.access);
     const authClaim = claims?.['https://api.openai.com/auth'];
     return authClaim?.chatgpt_plan_type;
@@ -81,7 +90,7 @@ function buildAuthorizeUrl(pkce, state) {
         response_type: 'code',
         client_id: CLIENT_ID,
         redirect_uri: REDIRECT_URI,
-        scope: 'openid profile email offline_access',
+        scope: 'openid profile email offline_access api.connectors.read api.connectors.invoke',
         code_challenge: pkce.challenge,
         code_challenge_method: 'S256',
         id_token_add_organizations: 'true',
@@ -236,6 +245,7 @@ export async function chatGptOAuthLogin() {
     const auth = {
         email: extractEmail(tokens),
         accountId: extractAccountId(tokens),
+        plan: extractPlanTypeFromTokens(tokens),
         refresh: tokens.refresh_token,
         access: tokens.access_token,
         expires: now + (tokens.expires_in ?? 3600) * 1000,
@@ -266,6 +276,7 @@ export async function refreshChatGptToken(auth) {
         const json = (await response.json());
         return {
             ...auth,
+            plan: extractPlanTypeFromTokens(json) ?? auth.plan,
             access: json.access_token,
             refresh: json.refresh_token ?? auth.refresh,
             expires: Date.now() + (json.expires_in ?? 3600) * 1000,
@@ -276,19 +287,15 @@ export async function refreshChatGptToken(auth) {
     }
 }
 /**
- * Get a valid access token from a ChatGptAuth object.
- * Auto-refreshes if expired. Returns the token string or an Error.
- * If the auth was refreshed, the `onRefresh` callback is called with
- * the updated auth so the caller can persist it.
+ * Refresh the auth object if needed and return the full updated state.
  */
-export async function getValidAccessToken(auth, onRefresh) {
-    // Token still valid (with 60s buffer)
+export async function getValidChatGptAuth(auth, onRefresh) {
     if (auth.expires > Date.now() + 60_000) {
-        return auth.access;
+        return auth;
     }
     const refreshed = await refreshChatGptToken(auth);
     if (refreshed instanceof Error)
         return refreshed;
     onRefresh?.(refreshed);
-    return refreshed.access;
+    return refreshed;
 }