some more clean up

caption.py

@@ -3,7 +3,7 @@ import io
 import os
 from together import Together
 
-MODEL = "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"
+MODEL = "deepseek-ai/DeepSeek-V3"
 TRIGGER_WORD = "tr1gger"
 
 def get_system_prompt():
@@ -106,7 +106,7 @@ def caption_single_image(client, img_str):
 
     # Request caption for the image using Llama 4 Maverick
     response = client.chat.completions.create(
-        model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
+        model=MODEL,
         messages=messages
     )
 

demo.py

@@ -504,8 +504,8 @@ def create_config_area():
         gr.Markdown("""
         **Note about categorization:**
         - Images are grouped by the part of the filename before the last underscore
-        - For example: "character_pose_01.png" and "character_pose_02.png" share the category "character_pose"
-        - When using "Batch by category", similar images are captioned together for more consistent results
+        - For example: "character_pose_1.png" and "character_pose_2.png" share the category "character_pose"
+        - When using "Batch process by category", similar images are captioned together for more consistent results
         """, elem_classes=["category-info"])
         
         caption_btn = gr.Button("Caption Images", variant="primary", interactive=False)

prompt.py

@@ -3,6 +3,7 @@ import argparse
 from pathlib import Path
 from caption import get_system_prompt, get_together_client, extract_captions
 
+MODEL = "deepseek-ai/DeepSeek-V3"
 
 def optimize_prompt(user_prompt, captions_dir=None, captions_list=None):
     """Optimize a user prompt to follow the same format as training captions.
@@ -43,7 +44,7 @@ def optimize_prompt(user_prompt, captions_dir=None, captions_list=None):
     ]
 
     response = client.chat.completions.create(
-        model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
+        model=MODEL,
         messages=messages
     )