feat(sdk): native OpenRouter audio/video/image routing across Python, TS, Go

sdk/go/ai/media_integration_test.go

@@ -1,6 +1,7 @@
 package ai
 
 import (
+	"bytes"
 	"context"
 	"encoding/base64"
 	"encoding/json"
@@ -250,7 +251,7 @@ func TestIntegrationAudioSSEStream(t *testing.T) {
 		var payload map[string]any
 		require.NoError(t, json.NewDecoder(r.Body).Decode(&payload))
 		assert.Equal(t, true, payload["stream"])
-		assert.Equal(t, "openai/gpt-4o-mini-tts", payload["model"])
+		assert.Equal(t, "openai/gpt-audio-mini", payload["model"])
 
 		w.Header().Set("Content-Type", "text/event-stream")
 		flusher, _ := w.(http.Flusher)
@@ -274,18 +275,21 @@ func TestIntegrationAudioSSEStream(t *testing.T) {
 		BaseURL: srv.URL,
 		Client:  srv.Client(),
 	}
+	// Pre-seed metadata so routing picks the chat-completions path
+	// without hitting the real OpenRouter `/models/.../endpoints` endpoint.
+	p.SeedModelMeta("openai/gpt-audio-mini", []string{"text", "audio"}, []string{"text"})
 
 	resp, err := p.GenerateAudio(context.Background(), AudioRequest{
 		Text:   "Say hello",
-		Model:  "openrouter/openai/gpt-4o-mini-tts",
+		Model:  "openrouter/openai/gpt-audio-mini",
 		Voice:  "nova",
-		Format: "wav",
+		Format: "mp3", // avoid pcm→wav rewrap so we can compare raw bytes
 	})
 
 	require.NoError(t, err)
 	assert.Equal(t, "Hello world", resp.Text)
 	require.NotNil(t, resp.Audio)
-	assert.Equal(t, "wav", resp.Audio.Format)
+	assert.Equal(t, "mp3", resp.Audio.Format)
 	assert.NotEmpty(t, resp.Audio.Data)
 
 	// Decode and verify audio bytes
@@ -315,16 +319,55 @@ func TestIntegrationAudioWithCustomFormat(t *testing.T) {
 		BaseURL: srv.URL,
 		Client:  srv.Client(),
 	}
+	p.SeedModelMeta("openai/gpt-audio-mini", []string{"text", "audio"}, []string{"text"})
 
 	resp, err := p.GenerateAudio(context.Background(), AudioRequest{
 		Text:   "test",
+		Model:  "openai/gpt-audio-mini",
 		Voice:  "echo",
 		Format: "mp3",
 	})
 	require.NoError(t, err)
 	assert.Equal(t, "mp3", resp.Audio.Format)
 }
 
+func TestIntegrationAudioSpeechEndpoint(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		assert.Equal(t, "/audio/speech", r.URL.Path)
+		assert.Equal(t, "Bearer test-key", r.Header.Get("Authorization"))
+
+		var payload map[string]any
+		require.NoError(t, json.NewDecoder(r.Body).Decode(&payload))
+		assert.Equal(t, "hexgrad/kokoro-82m", payload["model"])
+		assert.Equal(t, "af_bella", payload["voice"])
+		assert.Equal(t, "pcm", payload["response_format"]) // wav → pcm on wire
+
+		// Return 1KB of fake PCM16
+		pcm := bytes.Repeat([]byte{0x00, 0x01}, 500)
+		w.Header().Set("Content-Type", "audio/pcm")
+		w.WriteHeader(http.StatusOK)
+		_, _ = w.Write(pcm)
+	}))
+	defer srv.Close()
+
+	p := &OpenRouterMediaProvider{APIKey: "test-key", BaseURL: srv.URL, Client: srv.Client()}
+	p.SeedModelMeta("hexgrad/kokoro-82m", []string{"speech"}, []string{"text"})
+
+	resp, err := p.GenerateAudio(context.Background(), AudioRequest{
+		Text:   "hello",
+		Model:  "openrouter/hexgrad/kokoro-82m",
+		Voice:  "af_bella",
+		Format: "wav",
+	})
+	require.NoError(t, err)
+	require.NotNil(t, resp.Audio)
+	assert.Equal(t, "wav", resp.Audio.Format)
+	decoded, err := base64.StdEncoding.DecodeString(resp.Audio.Data)
+	require.NoError(t, err)
+	assert.Equal(t, []byte("RIFF"), decoded[:4])
+	assert.Equal(t, []byte("WAVE"), decoded[8:12])
+}
+
 func TestIntegrationImageGeneration(t *testing.T) {
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		var payload map[string]any

sdk/go/ai/media_provider.go

@@ -17,35 +17,69 @@ type VideoRequest struct {
 	AspectRatio     string           `json:"aspect_ratio,omitempty"`
 	GenerateAudio   *bool            `json:"generate_audio,omitempty"`
 	Seed            *int             `json:"seed,omitempty"`
+	// ImageURL is a single input image for image-to-video models (convenience
+	// alternative to FrameImages with frame_type=first_frame).
+	ImageURL        string           `json:"image_url,omitempty"`
+	// FrameImages is per-frame guidance — first_frame / last_frame. Items
+	// follow OpenRouter's shape: {type, image_url:{url}, frame_type}.
 	FrameImages     []map[string]any `json:"frame_images,omitempty"`
+	// InputReferences supplies style/subject reference images (Veo
+	// "reference-to-video").
 	InputReferences []map[string]any `json:"input_references,omitempty"`
 	PollInterval    time.Duration    `json:"-"`
 	Timeout         time.Duration    `json:"-"`
+	// Extra passes through additional model-specific parameters (e.g. Veo's
+	// `personGeneration`).
 	Extra           map[string]any   `json:"-"`
 }
 
 // ImageRequest holds parameters for image generation.
 type ImageRequest struct {
-	Prompt      string       `json:"prompt"`
-	Model       string       `json:"model,omitempty"`
-	Size        string       `json:"size,omitempty"`
-	Quality     string       `json:"quality,omitempty"`
-	ImageConfig *ImageConfig `json:"image_config,omitempty"`
+	Prompt      string         `json:"prompt"`
+	Model       string         `json:"model,omitempty"`
+	Size        string         `json:"size,omitempty"`
+	Quality     string         `json:"quality,omitempty"`
+	// ImageURLs are reference / source images for image+text→image models
+	// (e.g. x-ai/grok-imagine-image-quality). Each may be an http(s) or
+	// data: URL.
+	ImageURLs   []string       `json:"-"`
+	ImageConfig *ImageConfig   `json:"image_config,omitempty"`
+	// Extra passes through additional model-specific parameters.
+	Extra       map[string]any `json:"-"`
 }
 
 // ImageConfig holds OpenRouter-specific image configuration.
 type ImageConfig struct {
-	AspectRatio               string   `json:"aspect_ratio,omitempty"`
-	ImageSize                 string   `json:"image_size,omitempty"`
-	SuperResolutionReferences []string `json:"super_resolution_references,omitempty"`
+	AspectRatio               string     `json:"aspect_ratio,omitempty"`
+	ImageSize                 string     `json:"image_size,omitempty"`
+	// Strength is the image-to-image blend (0–1, model-dependent).
+	Strength                  *float64   `json:"strength,omitempty"`
+	// Style hint (e.g. Recraft V3 styles).
+	Style                     string     `json:"style,omitempty"`
+	// RgbColors is a color palette — array of [r,g,b].
+	RgbColors                 [][3]int   `json:"rgb_colors,omitempty"`
+	// BackgroundRgbColor is [r,g,b].
+	BackgroundRgbColor        *[3]int    `json:"background_rgb_color,omitempty"`
+	SuperResolutionReferences []string   `json:"super_resolution_references,omitempty"`
+	FontInputs                []FontInput `json:"font_inputs,omitempty"`
+}
+
+// FontInput configures custom text rendering for compatible image models.
+type FontInput struct {
+	FontURL string `json:"font_url"`
+	Text    string `json:"text"`
 }
 
 // AudioRequest holds parameters for audio generation.
 type AudioRequest struct {
-	Text   string `json:"text"`
-	Model  string `json:"model,omitempty"`
-	Voice  string `json:"voice,omitempty"`
-	Format string `json:"format,omitempty"`
+	Text   string         `json:"text"`
+	Model  string         `json:"model,omitempty"`
+	Voice  string         `json:"voice,omitempty"`
+	Format string         `json:"format,omitempty"`
+	// Speed multiplier (OpenAI TTS respects; other models ignore).
+	Speed  *float64       `json:"speed,omitempty"`
+	// Extra passes through additional model-specific parameters.
+	Extra  map[string]any `json:"-"`
 }
 
 // MediaResponse holds the result of a media generation call.

sdk/go/ai/media_provider_test.go

@@ -219,15 +219,18 @@ func TestOpenRouterGenerateAudio(t *testing.T) {
 		BaseURL: srv.URL,
 		Client:  srv.Client(),
 	}
+	p.SeedModelMeta("openai/gpt-audio-mini", []string{"text", "audio"}, []string{"text"})
 
 	resp, err := p.GenerateAudio(context.Background(), AudioRequest{
-		Text:  "Say hello",
-		Voice: "nova",
+		Text:   "Say hello",
+		Model:  "openai/gpt-audio-mini",
+		Voice:  "nova",
+		Format: "mp3",
 	})
 	require.NoError(t, err)
 	assert.Equal(t, "Hello", resp.Text)
 	require.NotNil(t, resp.Audio)
-	assert.Equal(t, "pcm16", resp.Audio.Format)
+	assert.Equal(t, "mp3", resp.Audio.Format)
 	assert.NotEmpty(t, resp.Audio.Data)
 }