Spaces:
Paused
Paused
Changed example prompts
Browse files
app.py
CHANGED
@@ -115,13 +115,13 @@ negPromptDesc = "This text will help deter the generation from converging toward
|
|
115 |
outputText = "This inferred imagery expands the field of view from the masked area of the input camera feed."
|
116 |
latentWalkDesc = "This allows you to walk short spans across the latent space with relatively continuous gradients."
|
117 |
|
118 |
-
examplePrompt1 = "A person in a room with colored hair"
|
119 |
-
examplePrompt2 = "People in a room with colored hair"
|
120 |
-
examplePrompt3 = "A man on a beach with long hair"
|
121 |
-
examplePrompt4 = "A woman on a beach with long hair"
|
122 |
-
examplePrompt5 = "A panda eating bamboo"
|
123 |
-
examplePrompt6 = "A family together in a room"
|
124 |
-
examplePrompt7 = "A family together outside with colored hair"
|
125 |
|
126 |
with gradio.Blocks(live=True) as ux:
|
127 |
gradio.Markdown("This generative machine learning demonstration streams stable diffusion outpainting inference live from your camera on your computer or phone to expand your local reality and create an alternate world. High quality frame to frame determinism is a hard problem to solve for latent diffusion models as the generation is inherently relative to input noise distributions for the latents, and many factors such as the inherent Bayer noise from the camera images as well as anything that is altered between camera images (such as focus, white balance, etc) causes non-determinism between frames. Some methods apply spationtemporal attention, but this demonstration focuses on the control over the input latents to navigate the latent space. **Increase the lighting of your physical scene from your camera's perspective, and avoid self shadows of scene content, to improve the quality and consistency of the scene generation.**")
|
|
|
115 |
outputText = "This inferred imagery expands the field of view from the masked area of the input camera feed."
|
116 |
latentWalkDesc = "This allows you to walk short spans across the latent space with relatively continuous gradients."
|
117 |
|
118 |
+
examplePrompt1 = "A person in a room" #A person in a room with colored hair"
|
119 |
+
examplePrompt2 = "A person with colored hair" #"People in a room with colored hair"
|
120 |
+
examplePrompt3 = "A person on a beach with long hair" #"A man on a beach with long hair"
|
121 |
+
examplePrompt4 = "A person in a field under a night sky" #"A woman on a beach with long hair"
|
122 |
+
examplePrompt5 = "A panda eating bamboo" #"A panda eating bamboo"
|
123 |
+
examplePrompt6 = "A bird flying in the sky" #"A family together in a room"
|
124 |
+
examplePrompt7 = "A Koala bear" #"A family together outside with colored hair"
|
125 |
|
126 |
with gradio.Blocks(live=True) as ux:
|
127 |
gradio.Markdown("This generative machine learning demonstration streams stable diffusion outpainting inference live from your camera on your computer or phone to expand your local reality and create an alternate world. High quality frame to frame determinism is a hard problem to solve for latent diffusion models as the generation is inherently relative to input noise distributions for the latents, and many factors such as the inherent Bayer noise from the camera images as well as anything that is altered between camera images (such as focus, white balance, etc) causes non-determinism between frames. Some methods apply spationtemporal attention, but this demonstration focuses on the control over the input latents to navigate the latent space. **Increase the lighting of your physical scene from your camera's perspective, and avoid self shadows of scene content, to improve the quality and consistency of the scene generation.**")
|