Update app.py

app.py

@@ -10,7 +10,7 @@ def create_captions_llava_llama3_docci(image):
     pipe = pipeline('Lin-Chen/open-llava-next-llama3-8b')
     gen_config = GenerationConfig(repetition_penalty=1.10)
     image = Image.fromarray(np.uint8(image)).convert('RGB')
-    response = pipe(('Provide a detailed, objective description of the visible content in the images for dataset annotations. Do not include any analysis, interpretations, explanations, or references to the purpose or themes of the image. Focus only on what is visually present', image), gen_config=gen_config)
+    response = pipe(('As an AI image annotation expert, please provide accurate annotations for these female photography images to enhance the CLIP model's understanding of the content. Prioritize tags by relevance. Your tag should include key elements such as the actions, attire, hairstyle, facial expressions, environment, dressing style, etc. of the characters in the image, as well as background content, and any other important tags. If the image has a distinct style or filter, it also needs to be labeled. If not, it does not need to be labeled separately. Starting with 1girl, there is no need to label woman. Your label should be accurate, non repetitive. These labels will be used for image reconstruction, so the closer the similarity to the original image, the better the label quality. Tags should be separated by commas. Special tags will receive a reward of $10 per image.', image), gen_config=gen_config)
     return response.text
 
 css = """