Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -23,36 +23,49 @@ def round_to_nearest_resolution_acceptable_by_vae(height, width):
|
|
| 23 |
height = height - (height % pipe.vae_temporal_compression_ratio)
|
| 24 |
width = width - (width % pipe.vae_temporal_compression_ratio)
|
| 25 |
return height, width
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
@spaces.GPU
|
| 28 |
def generate(prompt,
|
| 29 |
negative_prompt,
|
| 30 |
image,
|
| 31 |
video,
|
|
|
|
| 32 |
steps,
|
| 33 |
num_frames,
|
| 34 |
seed,
|
| 35 |
randomize_seed,
|
| 36 |
-
|
| 37 |
|
| 38 |
if randomize_seed:
|
| 39 |
seed = random.randint(0, MAX_SEED)
|
| 40 |
|
| 41 |
# Part 1. Generate video at smaller resolution
|
| 42 |
# Text-only conditioning is also supported without the need to pass `conditions`
|
| 43 |
-
expected_height, expected_width = 768, 1152
|
| 44 |
downscale_factor = 2 / 3
|
| 45 |
downscaled_height, downscaled_width = int(expected_height * downscale_factor), int(expected_width * downscale_factor)
|
| 46 |
downscaled_height, downscaled_width = round_to_nearest_resolution_acceptable_by_vae(downscaled_height, downscaled_width)
|
| 47 |
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
frames_to_use = 21
|
| 51 |
video = load_video(video)[:frames_to_use]
|
| 52 |
-
|
|
|
|
| 53 |
video = [image]
|
|
|
|
|
|
|
|
|
|
| 54 |
|
| 55 |
-
if condition
|
| 56 |
condition1 = LTXVideoCondition(video=video, frame_index=0)
|
| 57 |
latents = pipe(
|
| 58 |
conditions=condition1,
|
|
@@ -158,20 +171,19 @@ function refresh() {
|
|
| 158 |
with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
|
| 159 |
|
| 160 |
gr.Markdown("# LTX Video 0.9.7 Distilled")
|
| 161 |
-
|
| 162 |
with gr.Row():
|
| 163 |
with gr.Column():
|
| 164 |
with gr.Group():
|
| 165 |
-
with gr.Tab("text-to-video"):
|
| 166 |
image = gr.Image(label="", visible=False)
|
| 167 |
#prompt = gr.Textbox(label="prompt")
|
| 168 |
-
with gr.Tab("image-to-video"):
|
| 169 |
image = gr.Image(label="")
|
| 170 |
#prompt = gr.Textbox(label="prompt")
|
| 171 |
-
with gr.Tab("video-to-video"):
|
| 172 |
video = gr.Video(label="")
|
| 173 |
prompt = gr.Textbox(label="prompt")
|
| 174 |
-
t2v = gr.Checkbox(label="t2v", value=False)
|
| 175 |
run_button = gr.Button()
|
| 176 |
with gr.Column():
|
| 177 |
output = gr.Video(interactive=False)
|
|
@@ -184,20 +196,26 @@ with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
|
|
| 184 |
randomize_seed = gr.Checkbox(label="randomize seed")
|
| 185 |
with gr.Row():
|
| 186 |
steps = gr.Slider(label="Steps", minimum=1, maximum=30, value=8, step=1)
|
| 187 |
-
num_frames = gr.Slider(label="# frames", minimum=1, maximum=
|
| 188 |
|
| 189 |
|
| 190 |
-
|
|
|
|
|
|
|
|
|
|
| 191 |
run_button.click(fn=generate,
|
| 192 |
inputs=[prompt,
|
| 193 |
negative_prompt,
|
| 194 |
image,
|
| 195 |
-
video,
|
|
|
|
| 196 |
steps,
|
| 197 |
num_frames,
|
| 198 |
seed,
|
| 199 |
-
randomize_seed
|
| 200 |
outputs=[output])
|
|
|
|
|
|
|
| 201 |
|
| 202 |
|
| 203 |
demo.launch()
|
|
|
|
| 23 |
height = height - (height % pipe.vae_temporal_compression_ratio)
|
| 24 |
width = width - (width % pipe.vae_temporal_compression_ratio)
|
| 25 |
return height, width
|
| 26 |
+
|
| 27 |
+
def change_mode_to_text():
|
| 28 |
+
return gr.update(value="text-to-video")
|
| 29 |
+
|
| 30 |
+
def change_mode_to_image():
|
| 31 |
+
return gr.update(value="image-to-video")
|
| 32 |
+
|
| 33 |
+
def change_mode_to_video():
|
| 34 |
+
return gr.update(value="video-to-video")
|
| 35 |
|
| 36 |
@spaces.GPU
|
| 37 |
def generate(prompt,
|
| 38 |
negative_prompt,
|
| 39 |
image,
|
| 40 |
video,
|
| 41 |
+
mode,
|
| 42 |
steps,
|
| 43 |
num_frames,
|
| 44 |
seed,
|
| 45 |
randomize_seed,
|
| 46 |
+
improve_texture=False, progress=gr.Progress(track_tqdm=True)):
|
| 47 |
|
| 48 |
if randomize_seed:
|
| 49 |
seed = random.randint(0, MAX_SEED)
|
| 50 |
|
| 51 |
# Part 1. Generate video at smaller resolution
|
| 52 |
# Text-only conditioning is also supported without the need to pass `conditions`
|
| 53 |
+
expected_height, expected_width = 768, 1152 #todo make configurable
|
| 54 |
downscale_factor = 2 / 3
|
| 55 |
downscaled_height, downscaled_width = int(expected_height * downscale_factor), int(expected_width * downscale_factor)
|
| 56 |
downscaled_height, downscaled_width = round_to_nearest_resolution_acceptable_by_vae(downscaled_height, downscaled_width)
|
| 57 |
|
| 58 |
+
if mode == "text-to-video" and video is not None:
|
| 59 |
+
frames_to_use = 21 #todo make configurable
|
|
|
|
| 60 |
video = load_video(video)[:frames_to_use]
|
| 61 |
+
condition = True
|
| 62 |
+
elif mode == "image-to-video" and image is not None:
|
| 63 |
video = [image]
|
| 64 |
+
condition = True
|
| 65 |
+
else:
|
| 66 |
+
condition=False
|
| 67 |
|
| 68 |
+
if condition:
|
| 69 |
condition1 = LTXVideoCondition(video=video, frame_index=0)
|
| 70 |
latents = pipe(
|
| 71 |
conditions=condition1,
|
|
|
|
| 171 |
with gr.Blocks(css=css, theme=gr.themes.Ocean()) as demo:
|
| 172 |
|
| 173 |
gr.Markdown("# LTX Video 0.9.7 Distilled")
|
| 174 |
+
mode = gr.State(value="text-to-video")
|
| 175 |
with gr.Row():
|
| 176 |
with gr.Column():
|
| 177 |
with gr.Group():
|
| 178 |
+
with gr.Tab("text-to-video") as text_tab:
|
| 179 |
image = gr.Image(label="", visible=False)
|
| 180 |
#prompt = gr.Textbox(label="prompt")
|
| 181 |
+
with gr.Tab("image-to-video") as image_tab:
|
| 182 |
image = gr.Image(label="")
|
| 183 |
#prompt = gr.Textbox(label="prompt")
|
| 184 |
+
with gr.Tab("video-to-video") as video_tab:
|
| 185 |
video = gr.Video(label="")
|
| 186 |
prompt = gr.Textbox(label="prompt")
|
|
|
|
| 187 |
run_button = gr.Button()
|
| 188 |
with gr.Column():
|
| 189 |
output = gr.Video(interactive=False)
|
|
|
|
| 196 |
randomize_seed = gr.Checkbox(label="randomize seed")
|
| 197 |
with gr.Row():
|
| 198 |
steps = gr.Slider(label="Steps", minimum=1, maximum=30, value=8, step=1)
|
| 199 |
+
num_frames = gr.Slider(label="# frames", minimum=1, maximum=161, value=96, step=1)
|
| 200 |
|
| 201 |
|
| 202 |
+
text_tab.select(fn=change_mode_to_text, inputs=[], outputs=[mode])
|
| 203 |
+
image_tab.select(fn=change_mode_to_image, inputs=[], outputs=[mode])
|
| 204 |
+
video_tab.select(fn=change_mode_to_video, inputs=[], outputs=[mode])
|
| 205 |
+
|
| 206 |
run_button.click(fn=generate,
|
| 207 |
inputs=[prompt,
|
| 208 |
negative_prompt,
|
| 209 |
image,
|
| 210 |
+
video,
|
| 211 |
+
mode,
|
| 212 |
steps,
|
| 213 |
num_frames,
|
| 214 |
seed,
|
| 215 |
+
randomize_seed],
|
| 216 |
outputs=[output])
|
| 217 |
+
|
| 218 |
+
|
| 219 |
|
| 220 |
|
| 221 |
demo.launch()
|