Merge branch 'master' of https://git.yasue.org/ken/RandomVideoMaker

src/product/photo_download.ts

@@ -35,7 +35,7 @@ function extractJsonFromText(text: string): any | null {
     logger.info(`Starting photo download process with prompt: "${HARDCODED_USER_PROMPT}"`);
 
     // 1. Extract keywords from the hardcoded prompt
-    const keywords = ["food", "perfume", "accesory", "jewelry", "shoes", "bags", "watches", "sunglasses", "hats", "scarves", "belts", "wallets", "gloves", "ties", "cufflinks", "brooches", "necklaces", "bracelets", "earrings"];
+    const keywords = ["fullbody portrait girl", "fullbody portrait 18y girl", "fullbody portrait cute girl", "fullbody portrait blond girl", "fullbody portrait 20y girl"];
 
     if (!keywords || keywords.length === 0) {
         logger.error("Could not extract keywords from prompt. Exiting.");
@@ -48,7 +48,7 @@ function extractJsonFromText(text: string): any | null {
     for (const keyword of keywords) {
         try {
             logger.info(`Downloading photos for keyword: "${keyword}"`);
-            const downloadedPaths = await downloadImagesFromPinterestSearch(`${keyword} product photo`, PHOTOS_PER_KEYWORD);
+            const downloadedPaths = await downloadImagesFromPinterestSearch(`${keyword}`, PHOTOS_PER_KEYWORD);
             if (downloadedPaths.length > 0) {
                 logger.info(`Successfully downloaded ${downloadedPaths.length} images for "${keyword}"`);
                 totalDownloads += downloadedPaths.length;

src/tools/embed_prompt_to_png.ts

@@ -0,0 +1,54 @@
+import * as fs from 'fs';
+import * as path from 'path';
+import { callLMStudioAPIWithFile } from '../lib/lmstudio';
+import { embedJsonToPng } from '../lib/util';
+
+const imageDir = 'C:\\Users\\fm201\\Desktop\\vton\\bags';
+
+async function processImages() {
+    try {
+        const files = fs.readdirSync(imageDir);
+        const imageFiles = files.filter(file => /\.(png)$/i.test(file));
+
+        for (const imageFile of imageFiles) {
+            const imagePath = path.join(imageDir, imageFile);
+            console.log(`Processing ${imagePath}...`);
+
+            const prompt = `
+            Based on the handbag in the image, generate 10 outfit prompts that would complement it.
+            Each prompt should be a short, descriptive sentence of around 20 words.
+            Return the result in the following JSON format:
+            {"result": ["outfit prompt 1", "outfit prompt 2", ...]}
+            `;
+
+            try {
+                const response = await callLMStudioAPIWithFile(imagePath, prompt);
+                let outfitPrompts;
+
+                if (typeof response === 'string') {
+                    try {
+                        outfitPrompts = JSON.parse(response);
+                    } catch (e) {
+                        console.error(`Failed to parse JSON string for ${imageFile}:`, response);
+                        continue;
+                    }
+                } else {
+                    outfitPrompts = response;
+                }
+
+                if (outfitPrompts && outfitPrompts.result) {
+                    await embedJsonToPng(imagePath, outfitPrompts);
+                    console.log(`Successfully embedded prompts into ${imageFile}`);
+                } else {
+                    console.error(`Invalid JSON response for ${imageFile}:`, response);
+                }
+            } catch (error) {
+                console.error(`Failed to process ${imageFile}:`, error);
+            }
+        }
+    } catch (error) {
+        console.error('Error reading image directory:', error);
+    }
+}
+
+processImages();