feat(canvas): support multimodal video storyboard tables

backend/internal/app/logical_models_test.go

@@ -112,6 +112,14 @@ func TestMatchCapabilityRoutesTextImagesOnlyToDeclaredProfiles(t *testing.T) {
 	}
 }
 
+func TestMatchCapabilityRoutesUndeclaredVideoToTextMultimodalProfile(t *testing.T) {
+	profile := CapabilitySpec{Version: 1, Capability: "text"}
+	intent := ModelRequestIntent{Capability: "text", Inputs: map[string]int{"video": 1}}
+	if match := MatchCapability(profile, intent); !match.Matched {
+		t.Fatalf("multimodal text profile should accept legacy undeclared video input: %#v", match)
+	}
+}
+
 func TestMatchCapabilityAcceptsWildcardOptionValues(t *testing.T) {
 	spec := CapabilitySpec{
 		Version:    1,

backend/internal/app/model_router.go

@@ -274,9 +274,12 @@ func MatchCapability(spec CapabilitySpec, intent ModelRequestIntent) CapabilityM
 		}
 		constraint, declared := spec.Inputs[inputType]
 		if !declared {
-			// Multimodal text models may accept reference images even when the
-			// provider catalog does not explicitly declare the image input.
-			if count > 0 && !(normalizeCapability(spec.Capability) == "text" && inputType == "image") {
+			// Multimodal text models may accept reference images or videos even
+			// when the provider catalog does not explicitly declare the input.
+			// The concrete model capability still controls the frontend limits;
+			// this fallback keeps older catalog entries routable for known
+			// multimodal text endpoints.
+			if count > 0 && !(normalizeCapability(spec.Capability) == "text" && (inputType == "image" || inputType == "video")) {
 				reasons = append(reasons, "不支持 "+capabilityInputLabel(inputType)+"输入")
 			}
 			continue