diff --git a/tools/mtmd/mtmd-cli.cpp b/tools/mtmd/mtmd-cli.cpp
index 8ff93f08b9d..5e6cc79f379 100644
--- a/tools/mtmd/mtmd-cli.cpp
+++ b/tools/mtmd/mtmd-cli.cpp
@@ -228,10 +228,12 @@ static int eval_message(mtmd_cli_context & ctx, common_chat_msg & msg) {
     text.add_special   = add_bos;
     text.parse_special = true;
 
+    std::string formatted_chat;
+
     if (!mtmd_is_deepseekocr(ctx.ctx_vision.get())) {
-        auto formatted_chat = chat_add_and_format(ctx, msg);
+        formatted_chat = chat_add_and_format(ctx, msg);
         LOG_DBG("formatted_chat.prompt: %s\n", formatted_chat.c_str());
-        text.text = formatted_chat.c_str();        
+        text.text = formatted_chat.c_str();
     }
 
     if (g_is_interrupted) return 0;
@@ -316,8 +318,18 @@ int main(int argc, char ** argv) {
     if (is_single_turn) {
         g_is_generating = true;
         if (params.prompt.find(mtmd_default_marker()) == std::string::npos) {
-            for (size_t i = 0; i < params.image.size(); i++) {
-                params.prompt += mtmd_default_marker();
+            if (mtmd_is_deepseekocr(ctx.ctx_vision.get())) {
+                std::string image_tokens = "";
+                for (size_t i = 0; i < params.image.size(); i++) {
+                    image_tokens += mtmd_default_marker();
+                    image_tokens += '\n';
+                }
+                params.prompt = image_tokens + params.prompt;
+            }
+            else {
+                for (size_t i = 0; i < params.image.size(); i++) {
+                    params.prompt += mtmd_default_marker();
+                }
             }
         }
         common_chat_msg msg;