feat(gateway): add Seedream 5.0 Pro

2026-07-13 12:53:48 +08:00
parent 03abc0eab7
commit 9c300de72c
8 changed files with 460 additions and 5 deletions
@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/base64"
 	"encoding/json"
+	"fmt"
 	"net/http"
 	"net/http/httptest"
 	"strings"
@@ -1673,6 +1674,119 @@ func TestVolcesClientImageEditPreservesExplicitSequentialDisabledAndClampsMaxIma
 	}
 }

+func TestVolcesClientSeedreamProGenerationUsesRealModelAndSingleOutput(t *testing.T) {
+	var captured map[string]any
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if err := json.NewDecoder(r.Body).Decode(&captured); err != nil {
+			t.Fatalf("decode request: %v", err)
+		}
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"id":   "img-seedream-pro",
+			"data": []any{map[string]any{"url": "https://example.com/out.png"}},
+		})
+	}))
+	defer server.Close()
+
+	_, err := (VolcesClient{HTTPClient: server.Client()}).Run(context.Background(), Request{
+		Kind:      "images.generations",
+		ModelType: "image_generate",
+		Model:     "Seedream-5.0-Pro",
+		Body: map[string]any{
+			"model":                               "Seedream-5.0-Pro",
+			"prompt":                              "draw a mountain",
+			"resolution":                          "2K",
+			"n":                                   4,
+			"sequential_image_generation":         "auto",
+			"sequential_image_generation_options": map[string]any{"max_images": 4},
+		},
+		Candidate: store.RuntimeModelCandidate{
+			BaseURL:           server.URL,
+			ModelName:         "Seedream-5.0-Pro",
+			ModelAlias:        "Seedream-5.0-Pro",
+			ProviderModelName: "doubao-seedream-5-0-pro-260628",
+			Credentials:       map[string]any{"apiKey": "volces-key"},
+			Capabilities: map[string]any{
+				"image_generate": map[string]any{
+					"output_multiple_images":  false,
+					"output_max_images_count": 1,
+				},
+			},
+		},
+	})
+	if err != nil {
+		t.Fatalf("run Seedream Pro generation: %v", err)
+	}
+	if captured["model"] != "doubao-seedream-5-0-pro-260628" || captured["size"] != "2K" {
+		t.Fatalf("Seedream Pro should use the real model and resolution size, got %+v", captured)
+	}
+	if captured["n"] != float64(1) {
+		t.Fatalf("Seedream Pro should force single-image output, got %+v", captured)
+	}
+	if _, ok := captured["sequential_image_generation"]; ok {
+		t.Fatalf("Seedream Pro request must not include sequential image generation, got %+v", captured)
+	}
+	if _, ok := captured["sequential_image_generation_options"]; ok {
+		t.Fatalf("Seedream Pro request must not include sequential image options, got %+v", captured)
+	}
+}
+
+func TestVolcesClientSeedreamProEditForwardsTenReferencesAndCustomSize(t *testing.T) {
+	var captured map[string]any
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if err := json.NewDecoder(r.Body).Decode(&captured); err != nil {
+			t.Fatalf("decode request: %v", err)
+		}
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"id":   "img-seedream-pro-edit",
+			"data": []any{map[string]any{"url": "https://example.com/out.png"}},
+		})
+	}))
+	defer server.Close()
+
+	images := make([]any, 10)
+	for index := range images {
+		images[index] = fmt.Sprintf("https://example.com/reference-%d.png", index+1)
+	}
+	_, err := (VolcesClient{HTTPClient: server.Client()}).Run(context.Background(), Request{
+		Kind:      "images.edits",
+		ModelType: "image_edit",
+		Model:     "Seedream-5.0-Pro",
+		Body: map[string]any{
+			"model":  "Seedream-5.0-Pro",
+			"prompt": "combine the references",
+			"image":  images,
+			"width":  1600,
+			"height": 900,
+		},
+		Candidate: store.RuntimeModelCandidate{
+			BaseURL:           server.URL,
+			ModelName:         "Seedream-5.0-Pro",
+			ModelAlias:        "Seedream-5.0-Pro",
+			ProviderModelName: "doubao-seedream-5-0-pro-260628",
+			Credentials:       map[string]any{"apiKey": "volces-key"},
+			Capabilities: map[string]any{
+				"image_edit": map[string]any{
+					"input_multiple_images":     true,
+					"input_max_images_count":    10,
+					"output_multiple_images":    false,
+					"output_max_images_count":   1,
+					"input_max_file_size_bytes": 30 * 1024 * 1024,
+				},
+			},
+		},
+	})
+	if err != nil {
+		t.Fatalf("run Seedream Pro edit: %v", err)
+	}
+	gotImages, _ := captured["image"].([]any)
+	if captured["model"] != "doubao-seedream-5-0-pro-260628" || captured["size"] != "1600x900" || len(gotImages) != 10 {
+		t.Fatalf("Seedream Pro edit should preserve ten references and custom size, got %+v", captured)
+	}
+	if _, ok := captured["sequential_image_generation"]; ok {
+		t.Fatalf("Seedream Pro edit must not include sequential image generation, got %+v", captured)
+	}
+}
+
 func TestVolcesClientVideoSubmitsAndPollsTask(t *testing.T) {
 	var submitPath string
 	var pollPath string
@@ -781,14 +781,22 @@ func supportsMultipleOutputs(request Request, capabilityName string) bool {
 }

 func normalizeVolcesSequentialImageGeneration(body map[string]any, request Request) {
+	if !supportsMultipleOutputs(request, request.ModelType) {
+		delete(body, "sequential_image_generation")
+		delete(body, "sequential_image_generation_options")
+		if numericValue(body["n"], 1) > 1 {
+			body["n"] = 1
+		}
+		if numericValue(body["batch_size"], 1) > 1 {
+			body["batch_size"] = 1
+		}
+		return
+	}
 	options, hasOptions := volcesSequentialImageGenerationOptions(body)
 	normalizeVolcesSequentialMaxImages(options, request)
 	if _, explicit := body["sequential_image_generation"]; explicit {
 		return
 	}
-	if !supportsMultipleOutputs(request, request.ModelType) {
-		return
-	}
 	count := requestedVolcesSequentialImageCount(body, options)
 	if count <= 1 {
 		return