File size: 3,183 Bytes
812e69e
fed4c02
 
 
85f6fcb
fed4c02
 
1b24a66
fed4c02
 
 
812e69e
 
fed4c02
 
 
ab613df
69a9a62
ab613df
fed4c02
 
5d27053
fed4c02
 
5d27053
adefe82
bf7c515
 
fed4c02
bf7c515
 
ac6ae94
 
 
 
fed4c02
ac6ae94
fed4c02
bf7c515
 
fed4c02
 
 
 
 
 
 
 
 
 
4166d00
09fcd4a
c53b69e
fed4c02
ac6ae94
 
 
fed4c02
 
 
 
 
d55351a
 
 
 
fed4c02
 
 
 
 
 
 
 
 
 
 
 
 
 
ac6ae94
 
fed4c02
 
 
 
 
 
812e69e
fed4c02
 
 
 
 
 
55713b6
 
 
fed4c02
 
 
812e69e
 
fed4c02
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
import torch
from diffusers import UniPCMultistepScheduler, FlowMatchEulerDiscreteScheduler, DDIMScheduler, DPMSolverMultistepScheduler
from diffusers import WanPipeline, AutoencoderKLWan  # Use Wan-specific VAE
# from diffusers.hooks import apply_first_block_cache, FirstBlockCacheConfig
from para_attn.first_block_cache.diffusers_adapters import apply_cache_on_pipe
from diffusers.models import UNetSpatioTemporalConditionModel
from transformers import T5EncoderModel, T5Tokenizer
from huggingface_hub import hf_hub_download



from PIL import Image
import numpy as np



import gradio as gr
import spaces



device = "cuda" if torch.cuda.is_available() else "cpu"



model_id = "Wan-AI/Wan2.1-T2V-14B-Diffusers"
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=torch.bfloat16)
flow_shift = 1.0 #5.0  1.0 for image, 5.0 for 720P, 3.0 for 480P
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)

pipe.load_lora_weights(
    "joerose/Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32",
    weight_name="Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors"
)

pipe.fuse_lora()

pipe.to(device)



# print("Initialization complete. Gradio is starting...")









@spaces.GPU(size='xlarge')
def generate(prompt, negative_prompt, width=1024, height=1024, num_inference_steps=30, lora_id=None, progress=gr.Progress(track_tqdm=True)):
   if lora_id and lora_id.strip() != "":
       pipe.unload_lora_weights()
       pipe.load_lora_weights(lora_id.strip())



   #pipe.to("cuda")
   # apply_first_block_cache(pipe.transformer, FirstBlockCacheConfig(threshold=0.2))
   apply_cache_on_pipe(
      pipe,
       residual_diff_threshold=0.2,
   )
   try:
       output = pipe(
           prompt=prompt,
           negative_prompt=negative_prompt,
           height=height,
           width=width,
           num_frames=1,
           num_inference_steps=num_inference_steps,
           guidance_scale=1.0, #5.0
       )
       image = output.frames[0][0]
       image = (image * 255).astype(np.uint8)
       return Image.fromarray(image)
   finally:
       pipe.unload_lora_weights()
       # pass






iface = gr.Interface(
   fn=generate,
   inputs=[
       gr.Textbox(label="Input prompt"),
   ],
   additional_inputs = [
       gr.Textbox(label="Negative prompt", value = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards"),
       gr.Slider(label="Width", minimum=480, maximum=2048, step=16, value=1024),
       gr.Slider(label="Height", minimum=480, maximum=2048, step=16, value=1024),
       gr.Slider(minimum=1, maximum=80, step=1, label="Inference Steps", value=8),
       gr.Textbox(label="LoRA ID"),
   ],
   outputs=gr.Image(label="output"),
)



iface.launch()