sample with multiple text

examples/hunyuanvideo/hyvideo/config.py

@@ -323,7 +323,8 @@ def add_inference_args(parser: argparse.ArgumentParser):
         "--prompt",
         type=str,
         default=None,
-        help="Prompt for sampling during evaluation.",
+        help="A single prompt string or a path to a .txt file containing multiple prompts. "
+        "If a .txt file is provided, each line should contain one prompt.",
     )
     group.add_argument(
         "--seed-type",
@@ -365,11 +366,11 @@ def add_inference_args(parser: argparse.ArgumentParser):
     group.add_argument(
         "--text-embed-path",
         type=str,
-        help="path to npz containing text embeds, "
-        "including positive/negative prompt embed of text encoder 1 and 2"
-        ", and the mask for positive and negative prompt",
+        default=None,
+        help="A single .npz file path or a path to a .txt file containing multiple .npz file paths. "
+        "If a .txt file is provided, each line should contain one .npz file path. "
+        "This argument is required if `prompt` is a .txt file.",
     )
-
     # mindspore args
     group.add_argument("--ms-mode", type=int, default=0, help="0 graph, 1 pynative")
     group.add_argument(

examples/hunyuanvideo/hyvideo/inference.py

@@ -462,6 +462,11 @@ def predict(
         # ========================================================================
         # Arguments: prompt, new_prompt, negative_prompt
         # ========================================================================
+        if prompt is None:
+            if text_embed_path is None:
+                raise ValueError("Either `prompt` or `text_embed_path` must be specified.")
+            prompt = ""
+
         if not isinstance(prompt, str):
             raise TypeError(f"`prompt` must be a string, but got {type(prompt)}")
         prompt = [prompt.strip()]
@@ -532,6 +537,7 @@ def predict(
                          width: {target_width}
                   video_length: {target_video_length}
                         prompt: {prompt}
+               text_embed_path: {text_embed_path}
                     neg_prompt: {negative_prompt}
                           seed: {seed}
                    infer_steps: {infer_steps}

examples/hunyuanvideo/hyvideo/utils/file_utils.py

@@ -0,0 +1,52 @@
+def process_prompt_and_text_embed(prompt, text_embed_path):
+    """
+    Process the prompt and text embed path to ensure they are in the correct format and match in number.
+
+    Args:
+        prompt (str or None): A string prompt or path to a text file containing prompts.
+        text_embed_path (str or None): A path to a text file containing text embeddings or a npz file.
+
+    Returns:
+        tuple: A tuple containing two lists, the first with prompts and the second with text embed paths.
+
+    Raises:
+        ValueError: If the prompt and text embed path combination is invalid or the number of prompts and text embed paths do not match.
+    """
+    if prompt is None and text_embed_path is None:
+        raise ValueError("Either `prompt` or `text_embed_path` must be provided.")
+
+    if prompt is not None and prompt.endswith(".txt"):
+        with open(prompt, "r") as f:
+            prompts = [line.strip() for line in f.readlines()]
+        if text_embed_path is not None:
+            assert text_embed_path.endswith(
+                ".txt"
+            ), "When `prompt` is a txt file, `text_embed_path` should be a txt file too."
+            with open(text_embed_path, "r") as f:
+                text_embed_paths = [line.strip() for line in f.readlines()]
+        else:
+            text_embed_paths = [None] * len(prompts)
+    elif prompt is not None and isinstance(prompt, str):
+        prompts = [prompt.strip()]
+        if text_embed_path is not None:
+            assert text_embed_path.endswith(
+                ".npz"
+            ), "When `prompt` is a string, `text_embed_path` should be a npz file."
+            text_embed_paths = [text_embed_path.strip()]
+        else:
+            text_embed_paths = [None] * len(prompts)
+    elif text_embed_path is not None:
+        if text_embed_path.endswith(".txt"):
+            with open(text_embed_path, "r") as f:
+                text_embed_paths = [line.strip() for line in f.readlines()]
+            prompts = [None] * len(text_embed_paths)
+        else:
+            text_embed_paths = [text_embed_path.strip()]
+            prompts = [None]
+    else:
+        raise ValueError("Invalid combination of `prompt` and `text_embed_path`.")
+
+    if len(prompts) != len(text_embed_paths):
+        raise ValueError("The number of prompts and text embed paths must match.")
+
+    return prompts, text_embed_paths

examples/hunyuanvideo/sample_video.py

@@ -16,6 +16,7 @@
 sys.path.append(".")
 from hyvideo.config import parse_args
 from hyvideo.inference import HunyuanVideoSampler
+from hyvideo.utils.file_utils import process_prompt_and_text_embed
 from hyvideo.utils.ms_utils import init_env
 
 
@@ -27,9 +28,9 @@ def main():
         raise ValueError(f"`models_root` not exists: {models_root_path}")
 
     # Create save folder to save the samples
-    save_path = args.save_path if args.save_path_suffix == "" else f"{args.save_path}_{args.save_path_suffix}"
-    if not os.path.exists(args.save_path):
-        os.makedirs(save_path, exist_ok=True)
+    save_dir = args.save_path if args.save_path_suffix == "" else f"{args.save_path}_{args.save_path_suffix}"
+    if not os.path.exists(args.save_dir):
+        os.makedirs(save_dir, exist_ok=True)
 
     # ms env init
     rank_id, _ = init_env(
@@ -51,44 +52,54 @@ def main():
     args = hunyuan_video_sampler.args
 
     # Start sampling
-    # TODO: batch inference check
-    outputs = hunyuan_video_sampler.predict(
-        prompt=args.prompt,
-        height=args.video_size[0],
-        width=args.video_size[1],
-        video_length=args.video_length,
-        seed=args.seed,
-        negative_prompt=args.neg_prompt,
-        infer_steps=args.infer_steps,
-        guidance_scale=args.cfg_scale,
-        num_videos_per_prompt=args.num_videos,
-        flow_shift=args.flow_shift,
-        batch_size=args.batch_size,
-        embedded_guidance_scale=args.embedded_cfg_scale,
-        output_type=args.output_type,
-        text_embed_path=args.text_embed_path,
-    )
-    samples = outputs["samples"]
-
-    # Save samples
-    if rank_id == 0:
-        for i, sample in enumerate(samples):
-            sample = samples[i].unsqueeze(0)
-            time_flag = datetime.fromtimestamp(time.time()).strftime("%Y-%m-%d-%H:%M:%S")
-            save_path = (
-                f"{save_path}/{time_flag}_seed{outputs['seeds'][i]}_{outputs['prompts'][i][:100].replace('/','')}.mp4"
+    if args.prompt is None and args.text_embed_path is None:
+        raise ValueError("Either `prompt` or `text_embed_path` must be provided.")
+
+    prompts, text_embed_paths = process_prompt_and_text_embed(args.prompt, args.text_embed_path)
+    for prompt, text_embed_path in zip(prompts, text_embed_paths):
+        if prompt is not None:
+            logger.info(f"Sampling with prompt: {prompt}")
+        if text_embed_path is not None:
+            logger.info(f"Sampling with text embed path: {text_embed_path}")
+        try:
+            outputs = hunyuan_video_sampler.predict(
+                prompt=prompt,
+                height=args.video_size[0],
+                width=args.video_size[1],
+                video_length=args.video_length,
+                seed=args.seed,
+                negative_prompt=args.neg_prompt,
+                infer_steps=args.infer_steps,
+                guidance_scale=args.cfg_scale,
+                num_videos_per_prompt=args.num_videos,
+                flow_shift=args.flow_shift,
+                batch_size=args.batch_size,
+                embedded_guidance_scale=args.embedded_cfg_scale,
+                output_type=args.output_type,
+                text_embed_path=text_embed_path,
             )
+            samples = outputs["samples"]
+        except Exception as e:
+            logger.error(f"Error during prediction: {e}")
+            continue
+
+        # Save samples
+        if rank_id == 0:
+            for i, sample in enumerate(samples):
+                sample = samples[i].unsqueeze(0)
+                time_flag = datetime.fromtimestamp(time.time()).strftime("%Y-%m-%d-%H:%M:%S")
+                save_path = f"{save_dir}/{time_flag}_seed{outputs['seeds'][i]}_{outputs['prompts'][i][:100].replace('/','')}.mp4"
 
-            if args.output_type != "latent":
-                # save_videos_grid(sample, save_path, fps=24)
-                # b c t h w -> b t h w c
-                sample = sample.permute(0, 2, 3, 4, 1).asnumpy()
-                save_videos(sample, save_path, fps=24)
-            else:
-                save_path = save_path[:-4] + ".npy"
-                np.save(save_path, sample)
+                if args.output_type != "latent":
+                    # save_videos_grid(sample, save_path, fps=24)
+                    # b c t h w -> b t h w c
+                    sample = sample.permute(0, 2, 3, 4, 1).asnumpy()
+                    save_videos(sample, save_path, fps=24)
+                else:
+                    save_path = save_path[:-4] + ".npy"
+                    np.save(save_path, sample)
 
-            logger.info(f"Sample save to: {save_path}")
+                logger.info(f"Sample save to: {save_path}")
 
 
 if __name__ == "__main__":