diff --git a/modules/openai/test.py b/modules/openai/test.py
index 8c37104bb..34a3a06a8 100644
--- a/modules/openai/test.py
+++ b/modules/openai/test.py
@@ -20,7 +20,13 @@ model = AutoModelForCausalLM.from_pretrained(
     dtype=torch.bfloat16,
     trust_remote_code=True,
     attn_implementation="sdpa",
+    # attn_implementation="eager",
+    # attn_implementation="flash_attention_2",
 )
+model.config.use_flash_attention = True
+
+logger.log.info("OpenAI: eval model...")
+model.eval()
 tokenizer = AutoTokenizer.from_pretrained(
     "Qwen/Qwen3-0.6B",
     trust_remote_code=True
@@ -42,3 +48,4 @@ while True:
     except KeyboardInterrupt:
         server.stop()
         break
+
diff --git a/modules/video_models/video_prompt.py b/modules/video_models/video_prompt.py
index bbb29c409..963813e9d 100644
--- a/modules/video_models/video_prompt.py
+++ b/modules/video_models/video_prompt.py
@@ -1,7 +1,7 @@
 from modules import shared, extra_networks, ui_video_vlm
 
 
-def prepare_prompt(p, init_image, prompt:str, vlm_enhance:bool, vlm_model:str, vlm_system_prompt:str):
+def prepare_prompts(p, init_image, prompt:str, vlm_enhance:bool, vlm_model:str, vlm_system_prompt:str):
     p.prompt = shared.prompt_styles.apply_styles_to_prompt(p.prompt, p.styles)
     p.negative_prompt = shared.prompt_styles.apply_negative_styles_to_prompt(p.negative_prompt, p.styles)
     shared.prompt_styles.apply_styles_to_extra(p)
@@ -18,4 +18,7 @@ def prepare_prompt(p, init_image, prompt:str, vlm_enhance:bool, vlm_model:str, v
     )
     if new_prompt is not None and len(new_prompt) > 0:
         prompt = new_prompt
-    return prompt
+
+    p.styles = []
+    p.task_args['prompt'] = p.prompt
+    p.task_args['negative_prompt'] = p.negative_prompt
diff --git a/modules/video_models/video_run.py b/modules/video_models/video_run.py
index b23fd98e3..569220a45 100644
--- a/modules/video_models/video_run.py
+++ b/modules/video_models/video_run.py
@@ -95,12 +95,11 @@ def generate(*args, **kwargs):
     shared.sd_model = sd_models.apply_balanced_offload(shared.sd_model)
     devices.torch_gc(force=True, reason='video')
 
-    prompt = video_prompt.prepare_prompt(p, init_image, prompt, vlm_enhance, vlm_model, vlm_system_prompt)
 
     # set args
+    video_prompt.prepare_prompts(p, init_image, prompt, vlm_enhance, vlm_model, vlm_system_prompt)
     processing.fix_seed(p)
     video_vae.set_vae_params(p)
-    video_utils.set_prompt(p)
     p.task_args['num_inference_steps'] = p.steps
     p.task_args['width'] = p.width
     p.task_args['height'] = p.height
diff --git a/modules/video_models/video_utils.py b/modules/video_models/video_utils.py
index 054486648..18ae1249d 100644
--- a/modules/video_models/video_utils.py
+++ b/modules/video_models/video_utils.py
@@ -30,15 +30,6 @@ def check_av():
     return av
 
 
-def set_prompt(p):
-    p.prompt = shared.prompt_styles.apply_styles_to_prompt(p.prompt, p.styles)
-    p.negative_prompt = shared.prompt_styles.apply_negative_styles_to_prompt(p.negative_prompt, p.styles)
-    shared.prompt_styles.apply_styles_to_extra(p)
-    p.styles = []
-    p.task_args['prompt'] = p.prompt
-    p.task_args['negative_prompt'] = p.negative_prompt
-
-
 def hijack_encode_image(*args, **kwargs):
     t0 = time.time()
     try: