Base AI tag generation on existing tags

src/photo/actions.ts

@@ -9,6 +9,7 @@ import {
   getPhoto,
   getPhotos,
   addTagsToPhotos,
+  getUniqueTags,
 } from '@/photo/db/query';
 import { GetPhotosOptions, areOptionsSensitive } from './db';
 import {
@@ -37,7 +38,7 @@ import { blurImageFromUrl, extractImageDataFromBlobPath } from './server';
 import { TAG_FAVS, isTagFavs } from '@/tag';
 import { convertPhotoToPhotoDbInsert, Photo } from '.';
 import { runAuthenticatedAdminServerAction } from '@/auth';
-import { AI_IMAGE_QUERIES, AiImageQuery } from './ai';
+import { AiImageQuery, getAiImageQuery } from './ai';
 import { streamOpenAiImageQuery } from '@/services/openai';
 import {
   AI_TEXT_AUTO_GENERATED_FIELDS,
@@ -394,8 +395,13 @@ export const streamAiImageQueryAction = async (
   imageBase64: string,
   query: AiImageQuery,
 ) =>
-  runAuthenticatedAdminServerAction(() =>
-    streamOpenAiImageQuery(imageBase64, AI_IMAGE_QUERIES[query]));
+  runAuthenticatedAdminServerAction(async () => {
+    const existingTags = await getUniqueTags();
+    return streamOpenAiImageQuery(
+      imageBase64,
+      getAiImageQuery(query, existingTags),
+    );
+  });
 
 export const getImageBlurAction = async (url: string) =>
   runAuthenticatedAdminServerAction(() => blurImageFromUrl(url));

src/photo/ai/index.ts

@@ -1,5 +1,7 @@
 /* eslint-disable max-len */
 
+import { Tags } from '@/tag';
+
 export type AiAutoGeneratedField =
   'title' |
   'caption' |
@@ -42,15 +44,25 @@ export type AiImageQuery =
   'description-large' |
   'description-semantic';
 
-export const AI_IMAGE_QUERIES: Record<AiImageQuery, string> = {
-  'title': 'Write a compelling title for this image in 3 words or less',
-  'caption': 'Write a pithy caption for this image in 6 words or less and no punctuation',
-  'title-and-caption': 'Write a compelling title and pithy caption of 8 words or less for this image, using the format Title: "title" Caption: "caption"',
-  'tags': 'Describe this image three or less comma-separated keywords with no adjective or adverbs',
-  'description-small': 'Describe this image succinctly without the initial text "This image shows" or "This is a picture of"',
-  'description': 'Describe this image',
-  'description-large': 'Describe this image in detail',
-  'description-semantic': 'List up to 5 things in this image without description as a comma-separated list',
+export const getAiImageQuery = (
+  query: AiImageQuery,
+  existingTags: Tags = [],
+): string => {
+  switch (query) {
+  case 'title': return 'Write a compelling title for this image in 3 words or less';
+  case 'caption': return 'Write a pithy caption for this image in 6 words or less and no punctuation';
+  case 'title-and-caption': return 'Write a compelling title and pithy caption of 8 words or less for this image, using the format Title: "title" Caption: "caption"';
+  case 'tags':
+    const tagQuery = 'Describe this image three or less comma-separated keywords with no adjective or adverbs';
+    const tags = existingTags.map(({ tag }) => tag).join(', ');
+    return tags
+      ? `${tagQuery}. Consider using some of these existing tags, but only if they are relevant: ${tags}.`
+      : tagQuery;
+  case 'description-small': return 'Describe this image succinctly without the initial text "This image shows" or "This is a picture of"';
+  case 'description': return 'Describe this image';
+  case 'description-large': return 'Describe this image in detail';
+  case 'description-semantic': return 'List up to 5 things in this image without description as a comma-separated list';
+  }
 };
 
 export const parseTitleAndCaption = (text: string) => {

src/photo/ai/server.ts

@@ -1,9 +1,10 @@
 import { generateOpenAiImageQuery } from '@/services/openai';
 import {
-  AI_IMAGE_QUERIES,
   AiAutoGeneratedField,
+  getAiImageQuery,
   parseTitleAndCaption,
 } from '.';
+import { getUniqueTags } from '../db/query';
 
 export const generateAiImageQueries = async (
   imageBase64?: string,
@@ -29,7 +30,7 @@ export const generateAiImageQueries = async (
       ) {
         const titleAndCaption = await generateOpenAiImageQuery(
           imageBase64,
-          AI_IMAGE_QUERIES['title-and-caption'],
+          getAiImageQuery('title-and-caption'),
         );
         if (titleAndCaption) {
           const titleAndCaptionParsed = parseTitleAndCaption(titleAndCaption);
@@ -40,28 +41,29 @@ export const generateAiImageQueries = async (
         if (textFieldsToGenerate.includes('title')) {
           title = await generateOpenAiImageQuery(
             imageBase64,
-            AI_IMAGE_QUERIES['title'],
+            getAiImageQuery('title'),
           );
         }
         if (textFieldsToGenerate.includes('caption')) {
           caption =  await generateOpenAiImageQuery(
             imageBase64,
-            AI_IMAGE_QUERIES['caption'],
+            getAiImageQuery('caption'),
           );
         }
       }
   
       if (textFieldsToGenerate.includes('tags')) {
+        const existingTags = await getUniqueTags();
         tags = await generateOpenAiImageQuery(
           imageBase64,
-          AI_IMAGE_QUERIES['tags'],
+          getAiImageQuery('tags', existingTags),
         );
       }
   
       if (textFieldsToGenerate.includes('semantic')) {
         semanticDescription = await generateOpenAiImageQuery(
           imageBase64,
-          AI_IMAGE_QUERIES['description-small'],
+          getAiImageQuery('description-small'),
         );
       }
     }

src/services/openai.ts

@@ -73,7 +73,7 @@ export const streamOpenAiImageQuery = async (
 
   if (args) {
     (async () => {
-      const { textStream } = await streamText(args);
+      const { textStream } = streamText(args);
       for await (const delta of textStream) {
         stream.update(cleanUpAiTextResponse(delta));
       }