cleaned up a little

public/scripts/extensions/shared.js

@@ -30,24 +30,20 @@ export async function getMultimodalCaption(base64Img, prompt) {
         base64Img = await createThumbnail(base64Img, maxSide, maxSide, 'image/jpeg');
     }
 
-    const apiResult = extension_settings.caption.multimodal_api === 'google' ?
+    const isGoogle = extension_settings.caption.multimodal_api === 'google';
-        await fetch('/api/google/caption-image', {
+    const apiResult = await fetch(`/api/${isGoogle ? 'google' : 'openai'}/caption-image`, {
-            method: 'POST',
-            headers: getRequestHeaders(),
-            body: JSON.stringify({
-                image: base64Img,
-                prompt: prompt,
-            }),
-        })
-        : await fetch('/api/openai/caption-image', {
         method: 'POST',
         headers: getRequestHeaders(),
         body: JSON.stringify({
             image: base64Img,
             prompt: prompt,
+            ...(isGoogle
+                ? {}
+                : {
                     api: extension_settings.caption.multimodal_api || 'openai',
                     model: extension_settings.caption.multimodal_model || 'gpt-4-vision-preview',
                 }),
+        }),
     });
 
     if (!apiResult.ok) {

src/endpoints/google.js

@@ -14,7 +14,7 @@ router.post('/caption-image', jsonParser, async (request, response) => {
                 parts: [
                     { text: request.body.prompt },
                     { inlineData: {
-                        mimeType: 'image/png',
+                        mimeType: 'image/png', //jpg images seem to work fine even with this mimetype set?
                         data: request.body.image,
                     },
                     }],
@@ -40,8 +40,14 @@ router.post('/caption-image', jsonParser, async (request, response) => {
         }
 
         const data = await result.json();
+        console.log('Multimodal captioning response', data);
 
-        const caption = data?.candidates[0].content.parts[0].text;
+        const candidates = data?.candidates;
+        if(!candidates) {
+            return response.status(500).send('No candidates found, image was most likely filtered.');
+        }
+
+        const caption = candidates[0].content.parts[0].text;
         if (!caption) {
             return response.status(500).send('No caption found');
         }