Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -1,134 +1,127 @@
|
|
1 |
#!/usr/bin/env python
|
2 |
"""
|
3 |
-
Gradio demo for Wan2.1
|
|
|
|
|
4 |
"""
|
5 |
-
|
6 |
import os
|
7 |
-
# Persist HF cache between launches
|
8 |
-
os.environ["HF_HOME"] = "/mnt/data/huggingface"
|
9 |
-
|
10 |
-
import torch
|
11 |
import numpy as np
|
|
|
12 |
import gradio as gr
|
13 |
-
from PIL import Image
|
14 |
-
import torchvision.transforms.functional as TF
|
15 |
-
from transformers import CLIPVisionModel, CLIPImageProcessor
|
16 |
from diffusers import WanImageToVideoPipeline, AutoencoderKLWan
|
17 |
from diffusers.utils import export_to_video
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
-
# -----------------------------------------------------------------------------
|
20 |
-
# CONFIGURATION
|
21 |
-
# -----------------------------------------------------------------------------
|
22 |
-
MODEL_ID = "Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers"
|
23 |
-
DTYPE = torch.float16
|
24 |
-
MAX_AREA = 1280 * 720
|
25 |
-
DEFAULT_FRAMES = 81
|
26 |
-
|
27 |
-
# -----------------------------------------------------------------------------
|
28 |
-
# PIPELINE LOADING (ONCE)
|
29 |
-
# -----------------------------------------------------------------------------
|
30 |
def load_pipeline():
|
31 |
-
# 1)
|
32 |
-
|
33 |
MODEL_ID, subfolder="image_encoder", torch_dtype=torch.float32
|
34 |
)
|
35 |
-
# 2) VAE
|
36 |
vae = AutoencoderKLWan.from_pretrained(
|
37 |
MODEL_ID, subfolder="vae", torch_dtype=DTYPE
|
38 |
)
|
39 |
-
# 3) CLIPImageProcessor
|
40 |
-
|
41 |
-
"
|
42 |
)
|
43 |
-
# 4)
|
44 |
pipe = WanImageToVideoPipeline.from_pretrained(
|
45 |
MODEL_ID,
|
46 |
-
image_encoder=clip_encoder,
|
47 |
vae=vae,
|
48 |
-
|
|
|
49 |
torch_dtype=DTYPE,
|
50 |
-
device_map="balanced",
|
51 |
)
|
52 |
-
# 5) Slice the VAE to cut VRAM spikes
|
53 |
-
try:
|
54 |
-
pipe.vae.enable_slicing()
|
55 |
-
except AttributeError:
|
56 |
-
pass
|
57 |
return pipe
|
58 |
|
59 |
-
#
|
60 |
PIPE = load_pipeline()
|
61 |
|
62 |
-
|
63 |
-
#
|
64 |
-
#
|
65 |
def aspect_resize(img: Image.Image, max_area=MAX_AREA):
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
|
|
70 |
return img.resize((w, h), Image.LANCZOS), h, w
|
71 |
|
72 |
-
def center_crop_resize(img: Image.Image, h
|
|
|
73 |
ratio = max(w / img.width, h / img.height)
|
74 |
-
img
|
75 |
(round(img.width * ratio), round(img.height * ratio)),
|
76 |
Image.LANCZOS
|
77 |
)
|
78 |
return TF.center_crop(img, [h, w])
|
79 |
|
80 |
-
|
81 |
-
#
|
82 |
-
#
|
83 |
def generate(
|
84 |
first_frame: Image.Image,
|
85 |
-
last_frame:
|
86 |
-
prompt:
|
87 |
-
|
88 |
-
steps:
|
89 |
-
guidance:
|
90 |
-
num_frames:
|
91 |
-
seed:
|
92 |
-
fps:
|
93 |
-
progress=
|
94 |
):
|
95 |
-
#
|
96 |
if seed == -1:
|
97 |
seed = torch.seed()
|
98 |
gen = torch.Generator(device=PIPE.device).manual_seed(seed)
|
99 |
|
100 |
-
#
|
101 |
-
progress(0,
|
102 |
-
|
103 |
-
if last_frame.size !=
|
|
|
104 |
last_frame = center_crop_resize(last_frame, h, w)
|
105 |
|
106 |
-
#
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
# Run the pipeline
|
111 |
-
out = PIPE(
|
112 |
-
image=f0,
|
113 |
last_image=last_frame,
|
114 |
prompt=prompt,
|
115 |
-
negative_prompt=
|
116 |
height=h,
|
117 |
width=w,
|
118 |
num_frames=num_frames,
|
119 |
num_inference_steps=steps,
|
120 |
guidance_scale=guidance,
|
121 |
generator=gen,
|
122 |
-
callback=cb
|
123 |
)
|
124 |
|
125 |
-
#
|
126 |
-
|
|
|
|
|
|
|
|
|
127 |
return video_path, seed
|
128 |
|
129 |
-
|
130 |
-
#
|
131 |
-
#
|
132 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
133 |
gr.Markdown("## Wan2.1 FLF2V – First & Last Frame → Video")
|
134 |
|
@@ -136,25 +129,26 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
136 |
first_img = gr.Image(label="First frame", type="pil")
|
137 |
last_img = gr.Image(label="Last frame", type="pil")
|
138 |
|
139 |
-
|
140 |
-
|
141 |
|
142 |
with gr.Accordion("Advanced parameters", open=False):
|
143 |
-
steps = gr.Slider(10, 50, value=30, step=1,
|
144 |
-
guidance = gr.Slider(0.0, 10.0, value=5.5, step=0.1, label="Guidance")
|
145 |
num_frames = gr.Slider(16, 129, value=DEFAULT_FRAMES, step=1, label="Frames")
|
146 |
-
fps = gr.Slider(4, 30, value=16, step=1,
|
147 |
-
|
148 |
|
149 |
-
run_btn
|
150 |
-
|
151 |
-
|
152 |
|
153 |
run_btn.click(
|
154 |
fn=generate,
|
155 |
-
inputs=[
|
156 |
-
|
157 |
-
outputs=[
|
158 |
)
|
159 |
|
160 |
-
|
|
|
|
1 |
#!/usr/bin/env python
|
2 |
"""
|
3 |
+
Gradio demo for Wan2.1 First-Last-Frame-to-Video (FLF2V)
|
4 |
+
Loads the huge model once, uses balanced device placement,
|
5 |
+
streams high-level progress, and auto-offers the .mp4 for download.
|
6 |
"""
|
|
|
7 |
import os
|
|
|
|
|
|
|
|
|
8 |
import numpy as np
|
9 |
+
import torch
|
10 |
import gradio as gr
|
|
|
|
|
|
|
11 |
from diffusers import WanImageToVideoPipeline, AutoencoderKLWan
|
12 |
from diffusers.utils import export_to_video
|
13 |
+
from transformers import CLIPImageProcessor, CLIPVisionModel
|
14 |
+
from PIL import Image
|
15 |
+
import torchvision.transforms.functional as TF
|
16 |
+
|
17 |
+
# --------------------------------------------------------------------
|
18 |
+
# CONFIG
|
19 |
+
MODEL_ID = "Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers"
|
20 |
+
DTYPE = torch.float16 # half-precision
|
21 |
+
MAX_AREA = 1280 * 720 # ≤720p
|
22 |
+
DEFAULT_FRAMES = 81 # ≈5s @16fps
|
23 |
+
# --------------------------------------------------------------------
|
24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
def load_pipeline():
|
26 |
+
# 1) image encoder in full precision
|
27 |
+
image_encoder = CLIPVisionModel.from_pretrained(
|
28 |
MODEL_ID, subfolder="image_encoder", torch_dtype=torch.float32
|
29 |
)
|
30 |
+
# 2) VAE in reduced precision
|
31 |
vae = AutoencoderKLWan.from_pretrained(
|
32 |
MODEL_ID, subfolder="vae", torch_dtype=DTYPE
|
33 |
)
|
34 |
+
# 3) CLIPImageProcessor so we get the right class
|
35 |
+
image_processor = CLIPImageProcessor.from_pretrained(
|
36 |
+
MODEL_ID, subfolder="", torch_dtype=DTYPE
|
37 |
)
|
38 |
+
# 4) load everything with a balanced device map
|
39 |
pipe = WanImageToVideoPipeline.from_pretrained(
|
40 |
MODEL_ID,
|
|
|
41 |
vae=vae,
|
42 |
+
image_encoder=image_encoder,
|
43 |
+
image_processor=image_processor,
|
44 |
torch_dtype=DTYPE,
|
45 |
+
device_map="balanced", # splits weights CPU/GPU
|
46 |
)
|
|
|
|
|
|
|
|
|
|
|
47 |
return pipe
|
48 |
|
49 |
+
# load once at import
|
50 |
PIPE = load_pipeline()
|
51 |
|
52 |
+
|
53 |
+
# --------------------------------------------------------------------
|
54 |
+
# UTILS
|
55 |
def aspect_resize(img: Image.Image, max_area=MAX_AREA):
|
56 |
+
"""Resize while respecting multiples of the model’s patch size."""
|
57 |
+
ar = img.height / img.width
|
58 |
+
mod = PIPE.vae_scale_factor_spatial * PIPE.transformer.config.patch_size[1]
|
59 |
+
h = round(np.sqrt(max_area * ar)) // mod * mod
|
60 |
+
w = round(np.sqrt(max_area / ar)) // mod * mod
|
61 |
return img.resize((w, h), Image.LANCZOS), h, w
|
62 |
|
63 |
+
def center_crop_resize(img: Image.Image, h, w):
|
64 |
+
"""Crop-and-resize to exactly (h, w)."""
|
65 |
ratio = max(w / img.width, h / img.height)
|
66 |
+
img = img.resize(
|
67 |
(round(img.width * ratio), round(img.height * ratio)),
|
68 |
Image.LANCZOS
|
69 |
)
|
70 |
return TF.center_crop(img, [h, w])
|
71 |
|
72 |
+
|
73 |
+
# --------------------------------------------------------------------
|
74 |
+
# GENERATE (with simple progress streaming)
|
75 |
def generate(
|
76 |
first_frame: Image.Image,
|
77 |
+
last_frame: Image.Image,
|
78 |
+
prompt: str,
|
79 |
+
negative_prompt: str,
|
80 |
+
steps: int,
|
81 |
+
guidance: float,
|
82 |
+
num_frames: int,
|
83 |
+
seed: int,
|
84 |
+
fps: int,
|
85 |
+
progress=gr.Progress(), # gradio’s built-in progress callback
|
86 |
):
|
87 |
+
# pick or set seed
|
88 |
if seed == -1:
|
89 |
seed = torch.seed()
|
90 |
gen = torch.Generator(device=PIPE.device).manual_seed(seed)
|
91 |
|
92 |
+
# 0→10%: resize
|
93 |
+
progress(0.0, desc="Resizing first frame…")
|
94 |
+
first_frame, h, w = aspect_resize(first_frame)
|
95 |
+
if last_frame.size != first_frame.size:
|
96 |
+
progress(0.1, desc="Resizing last frame…")
|
97 |
last_frame = center_crop_resize(last_frame, h, w)
|
98 |
|
99 |
+
# 10→20%: ready to run
|
100 |
+
progress(0.2, desc="Starting video inference…")
|
101 |
+
result = PIPE(
|
102 |
+
image=first_frame,
|
|
|
|
|
|
|
103 |
last_image=last_frame,
|
104 |
prompt=prompt,
|
105 |
+
negative_prompt=negative_prompt or None,
|
106 |
height=h,
|
107 |
width=w,
|
108 |
num_frames=num_frames,
|
109 |
num_inference_steps=steps,
|
110 |
guidance_scale=guidance,
|
111 |
generator=gen,
|
|
|
112 |
)
|
113 |
|
114 |
+
# 80→100%: export
|
115 |
+
progress(0.8, desc="Assembling video file…")
|
116 |
+
video_path = export_to_video(result.frames[0], fps=fps)
|
117 |
+
progress(1.0, desc="Done!")
|
118 |
+
|
119 |
+
# return path so gr.File offers immediate download, plus seed used
|
120 |
return video_path, seed
|
121 |
|
122 |
+
|
123 |
+
# --------------------------------------------------------------------
|
124 |
+
# UI
|
125 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
126 |
gr.Markdown("## Wan2.1 FLF2V – First & Last Frame → Video")
|
127 |
|
|
|
129 |
first_img = gr.Image(label="First frame", type="pil")
|
130 |
last_img = gr.Image(label="Last frame", type="pil")
|
131 |
|
132 |
+
prompt = gr.Textbox(label="Prompt", placeholder="A blue bird takes off…")
|
133 |
+
negative = gr.Textbox(label="Negative prompt (optional)", placeholder="ugly, blurry")
|
134 |
|
135 |
with gr.Accordion("Advanced parameters", open=False):
|
136 |
+
steps = gr.Slider(10, 50, value=30, step=1, label="Sampling steps")
|
137 |
+
guidance = gr.Slider(0.0, 10.0, value=5.5, step=0.1, label="Guidance scale")
|
138 |
num_frames = gr.Slider(16, 129, value=DEFAULT_FRAMES, step=1, label="Frames")
|
139 |
+
fps = gr.Slider(4, 30, value=16, step=1, label="FPS")
|
140 |
+
seed = gr.Number(value=-1, precision=0, label="Seed (-1=random)")
|
141 |
|
142 |
+
run_btn = gr.Button("Generate")
|
143 |
+
download = gr.File(label="Download video", interactive=False)
|
144 |
+
used_seed = gr.Number(label="Seed used", interactive=False)
|
145 |
|
146 |
run_btn.click(
|
147 |
fn=generate,
|
148 |
+
inputs=[first_img, last_img, prompt, negative,
|
149 |
+
steps, guidance, num_frames, seed, fps],
|
150 |
+
outputs=[download, used_seed],
|
151 |
)
|
152 |
|
153 |
+
# queue tasks so users see the little task-queue progress bar
|
154 |
+
demo.queue().launch(server_name="0.0.0.0", server_port=7860)
|