Spaces:
Running
Running
Upload 3 files
Browse files- i2v_app_t4.py +12 -1
- t2v_app_t4.py +2 -2
- v2v_app_t4.py +12 -1
i2v_app_t4.py
CHANGED
@@ -123,6 +123,7 @@ def infer(
|
|
123 |
if seed == -1:
|
124 |
seed = random.randint(0, 2**8 - 1)
|
125 |
|
|
|
126 |
pipe_image = CogVideoXImageToVideoPipeline.from_pretrained(
|
127 |
"THUDM/CogVideoX-5b-I2V",
|
128 |
transformer=transformer,
|
@@ -132,6 +133,16 @@ def infer(
|
|
132 |
text_encoder=text_encoder,
|
133 |
torch_dtype=torch.float16
|
134 |
).to(device)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
image_input = Image.fromarray(image_input).resize(size=(720, 480)) # Convert to PIL
|
136 |
image = load_image(image_input)
|
137 |
video_pt = pipe_image(
|
@@ -144,7 +155,7 @@ def infer(
|
|
144 |
guidance_scale=7.0,
|
145 |
generator=torch.Generator(device="cpu").manual_seed(seed),
|
146 |
).frames
|
147 |
-
pipe_image.to("cpu")
|
148 |
del pipe_image
|
149 |
gc.collect()
|
150 |
torch.cuda.empty_cache()
|
|
|
123 |
if seed == -1:
|
124 |
seed = random.randint(0, 2**8 - 1)
|
125 |
|
126 |
+
'''
|
127 |
pipe_image = CogVideoXImageToVideoPipeline.from_pretrained(
|
128 |
"THUDM/CogVideoX-5b-I2V",
|
129 |
transformer=transformer,
|
|
|
133 |
text_encoder=text_encoder,
|
134 |
torch_dtype=torch.float16
|
135 |
).to(device)
|
136 |
+
'''
|
137 |
+
pipe_image = CogVideoXImageToVideoPipeline.from_pretrained(
|
138 |
+
"THUDM/CogVideoX-5b-I2V",
|
139 |
+
transformer=transformer,
|
140 |
+
vae=vae,
|
141 |
+
scheduler=pipe.scheduler,
|
142 |
+
tokenizer=pipe.tokenizer,
|
143 |
+
text_encoder=text_encoder,
|
144 |
+
torch_dtype=torch.float16
|
145 |
+
)
|
146 |
image_input = Image.fromarray(image_input).resize(size=(720, 480)) # Convert to PIL
|
147 |
image = load_image(image_input)
|
148 |
video_pt = pipe_image(
|
|
|
155 |
guidance_scale=7.0,
|
156 |
generator=torch.Generator(device="cpu").manual_seed(seed),
|
157 |
).frames
|
158 |
+
#pipe_image.to("cpu")
|
159 |
del pipe_image
|
160 |
gc.collect()
|
161 |
torch.cuda.empty_cache()
|
t2v_app_t4.py
CHANGED
@@ -121,7 +121,7 @@ def infer(
|
|
121 |
if seed == -1:
|
122 |
seed = random.randint(0, 2**8 - 1)
|
123 |
|
124 |
-
pipe.to(device)
|
125 |
video_pt = pipe(
|
126 |
prompt=prompt,
|
127 |
num_videos_per_prompt=1,
|
@@ -132,7 +132,7 @@ def infer(
|
|
132 |
guidance_scale=7.0,
|
133 |
generator=torch.Generator(device="cpu").manual_seed(seed),
|
134 |
).frames
|
135 |
-
pipe.to("cpu")
|
136 |
gc.collect()
|
137 |
torch.cuda.empty_cache()
|
138 |
return (video_pt, seed)
|
|
|
121 |
if seed == -1:
|
122 |
seed = random.randint(0, 2**8 - 1)
|
123 |
|
124 |
+
#pipe.to(device)
|
125 |
video_pt = pipe(
|
126 |
prompt=prompt,
|
127 |
num_videos_per_prompt=1,
|
|
|
132 |
guidance_scale=7.0,
|
133 |
generator=torch.Generator(device="cpu").manual_seed(seed),
|
134 |
).frames
|
135 |
+
#pipe.to("cpu")
|
136 |
gc.collect()
|
137 |
torch.cuda.empty_cache()
|
138 |
return (video_pt, seed)
|
v2v_app_t4.py
CHANGED
@@ -127,6 +127,7 @@ def infer(
|
|
127 |
seed = random.randint(0, 2**8 - 1)
|
128 |
|
129 |
video = load_video(video_input)[:49] # Limit to 49 frames
|
|
|
130 |
pipe_video = CogVideoXVideoToVideoPipeline.from_pretrained(
|
131 |
"THUDM/CogVideoX-5b",
|
132 |
transformer=transformer,
|
@@ -136,6 +137,16 @@ def infer(
|
|
136 |
text_encoder=text_encoder,
|
137 |
torch_dtype=torch.float16
|
138 |
).to(device)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
139 |
video_pt = pipe_video(
|
140 |
video=video,
|
141 |
prompt=prompt,
|
@@ -147,7 +158,7 @@ def infer(
|
|
147 |
guidance_scale=7.0,
|
148 |
generator=torch.Generator(device="cpu").manual_seed(seed),
|
149 |
).frames
|
150 |
-
pipe_video.to("cpu")
|
151 |
del pipe_video
|
152 |
gc.collect()
|
153 |
torch.cuda.empty_cache()
|
|
|
127 |
seed = random.randint(0, 2**8 - 1)
|
128 |
|
129 |
video = load_video(video_input)[:49] # Limit to 49 frames
|
130 |
+
'''
|
131 |
pipe_video = CogVideoXVideoToVideoPipeline.from_pretrained(
|
132 |
"THUDM/CogVideoX-5b",
|
133 |
transformer=transformer,
|
|
|
137 |
text_encoder=text_encoder,
|
138 |
torch_dtype=torch.float16
|
139 |
).to(device)
|
140 |
+
'''
|
141 |
+
pipe_video = CogVideoXVideoToVideoPipeline.from_pretrained(
|
142 |
+
"THUDM/CogVideoX-5b",
|
143 |
+
transformer=transformer,
|
144 |
+
vae=vae,
|
145 |
+
scheduler=pipe.scheduler,
|
146 |
+
tokenizer=pipe.tokenizer,
|
147 |
+
text_encoder=text_encoder,
|
148 |
+
torch_dtype=torch.float16
|
149 |
+
)
|
150 |
video_pt = pipe_video(
|
151 |
video=video,
|
152 |
prompt=prompt,
|
|
|
158 |
guidance_scale=7.0,
|
159 |
generator=torch.Generator(device="cpu").manual_seed(seed),
|
160 |
).frames
|
161 |
+
#pipe_video.to("cpu")
|
162 |
del pipe_video
|
163 |
gc.collect()
|
164 |
torch.cuda.empty_cache()
|