From ed9bfa9a54e957ceb2538118fd5ff53e49c4bdaf Mon Sep 17 00:00:00 2001
From: shrimbly <falloon.willie@gmail.com>
Date: Fri, 9 Jan 2026 20:25:55 +1300
Subject: [PATCH] fix(03-01): improve Replicate capability inference to exclude
 video models

Video models were getting text-to-image capability by default.
Now properly detects video models (by keywords like video, animate,
motion, luma, kling, minimax) and only assigns video capabilities.
---
 src/app/api/models/route.ts | 42 ++++++++++++++++++++++++-------------
 1 file changed, 27 insertions(+), 15 deletions(-)

diff --git a/src/app/api/models/route.ts b/src/app/api/models/route.ts
index 74a63be0..760a1d9e 100644
--- a/src/app/api/models/route.ts
+++ b/src/app/api/models/route.ts
@@ -138,32 +138,44 @@ type ModelsResponse = ModelsSuccessResponse | ModelsErrorResponse;
 // ============ Replicate Helpers ============
 
 function inferReplicateCapabilities(model: ReplicateModel): ModelCapability[] {
-  const capabilities: ModelCapability[] = ["text-to-image"];
-
+  const capabilities: ModelCapability[] = [];
   const searchText = `${model.name} ${model.description ?? ""}`.toLowerCase();
 
-  if (
-    searchText.includes("img2img") ||
-    searchText.includes("image-to-image") ||
-    searchText.includes("inpaint") ||
-    searchText.includes("controlnet")
-  ) {
-    capabilities.push("image-to-image");
-  }
-
-  if (
+  // Check for video-related keywords first
+  const isVideoModel =
     searchText.includes("video") ||
     searchText.includes("animate") ||
-    searchText.includes("motion")
-  ) {
+    searchText.includes("motion") ||
+    searchText.includes("luma") ||
+    searchText.includes("kling") ||
+    searchText.includes("minimax");
+
+  if (isVideoModel) {
+    // Video model - determine video capability type
     if (
       searchText.includes("img2vid") ||
-      searchText.includes("image-to-video")
+      searchText.includes("image-to-video") ||
+      searchText.includes("i2v")
     ) {
       capabilities.push("image-to-video");
     } else {
       capabilities.push("text-to-video");
     }
+  } else {
+    // Image model - default to text-to-image
+    capabilities.push("text-to-image");
+
+    // Check for image-to-image capability
+    if (
+      searchText.includes("img2img") ||
+      searchText.includes("image-to-image") ||
+      searchText.includes("inpaint") ||
+      searchText.includes("controlnet") ||
+      searchText.includes("upscale") ||
+      searchText.includes("restore")
+    ) {
+      capabilities.push("image-to-image");
+    }
   }
 
   return capabilities;