Add fps to pipe in wans2v

[hrudaykolla]

As of now fps of out video is hardcoaded as 16 from model_config.
Instead as cfg_scale and num_frames can be set. same way fps to be added.

Is there any reason fps is hardcoded?

[akaitsuki-ii]

@hrudaykolla In fact fps is not really hard coded. Such as at the end of this example, it's OK to call save_video with any provided fps, and we use fps of the pipeline by default, which is automatically adapted to different model loaded.

[hrudaykolla]

Thanks for responding, but what i was trying to convey is below in detail.
I was talking about examples/wan_speech_to_video.py.
for example there is no option in pipe to set fps, explanation below. When you run this now, the fps of video is 16.
'''
def wan_rs2v(pipe: WanSpeech2VideoPipeline, input_data_dir: str):
audio_path = f"{input_data_dir}/sing.mp3"
audio = librosa.load(audio_path, sr=16000)[0]
audio = torch.from_numpy(audio)[None] # (1, audio_len)
frames = pipe(
ref_image=Image.open(f"{input_data_dir}/woman.png").convert("RGB"),
audio=audio,
prompt="画面清晰，视频中，一个女人正在唱歌，表情动作十分投入",
negative_prompt="画面模糊，最差质量，画面模糊，细节模糊不清，情绪激动剧烈，手快速抖动，字幕，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
cfg_scale=4.5,
num_inference_steps=40,
seed=42,
num_frames_per_clip=80,
num_clips=3,
ref_as_first_frame=True,
)
save_video_with_audio(frames, audio_path=audio_path, target_video_path="wan_rs2v.mp4")
'''
diffsynth_engine/pipelines/wan_s2v.py , line 646,
'''config.fps = model_config.pop("fps", 16)'''

https://github.com/modelscope/DiffSynth-Engine/blob/main/diffsynth_engine/conf/models/wan/dit/wan2.2_s2v_14b.json
There is no fps in the model config. So it is being to set 16 by default.

What I was saying is, in https://github.com/modelscope/DiffSynth-Engine/blob/main/diffsynth_engine/pipelines/wan_s2v.py
Like the lines 439-442 for cfg_scale, num_inference_steps, ...
cfg_scale = self.config.cfg_scale if cfg_scale is None else cfg_scale
num_inference_steps = self.config.num_inference_steps if num_inference_steps is None else num_inference_steps
original_height, original_width = ref_image.height, ref_image.width

similarly an option for fps to be included, so that user can pass it through pipe.

I have updated this for me, I was curious why it was not set earlier as an option in pipe.