Spaces:
Sleeping
Sleeping
File size: 4,425 Bytes
e4b3d88 0a05d1f b2e9e1e 0a05d1f e4b3d88 0a05d1f e4b3d88 0a05d1f 9b68ef6 0a05d1f e4b3d88 0a05d1f e4b3d88 |
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 |
import gradio as gr
from PIL import Image, ImageColor
#my back_task code being imported
from back_task import *
# The function that does the hard work
def generate(radio,color,prompt, guidance_loss_scale):
print(color)
if radio == "color guidance":
target_color = ImageColor.getcolor(color, "RGB") # Target color as RGB
target_color = [a / 255 for a in target_color] # Rescale from (0, 255) to (0, 1)
elif radio == "text guidance":
# We embed a prompt with CLIP as our target
text = open_clip.tokenize([prompt]).to(device)
with torch.no_grad(), torch.cuda.amp.autocast():
text_features = clip_model.encode_text(text)
x = torch.randn(1, 3, 256, 256).to(device)
for i, t in tqdm(enumerate(scheduler.timesteps)):
model_input = scheduler.scale_model_input(x, t)
with torch.no_grad():
noise_pred = image_pipe.unet(model_input, t)["sample"]
if radio == "color guidance":
x = x.detach().requires_grad_()
x0 = scheduler.step(noise_pred, t, x).pred_original_sample
loss = color_loss(x0, target_color) * guidance_loss_scale
cond_grad = -torch.autograd.grad(loss, x)[0]
x = x.detach() + cond_grad
elif radio == "text guidance":
cond_grad = 0
for cut in range(n_cuts):
# Set requires grad on x
x = x.detach().requires_grad_()
# Get the predicted x0:
x0 = scheduler.step(noise_pred, t, x).pred_original_sample
# Calculate loss
loss = clip_loss(x0, text_features) * guidance_loss_scale
# Get gradient (scale by n_cuts since we want the average)
cond_grad -= torch.autograd.grad(loss, x)[0] / n_cuts
# Modify x based on this gradient
alpha_bar = scheduler.alphas_cumprod[i]
x = x.detach() + cond_grad * alpha_bar.sqrt() # Note the additional scaling factor here!
x = scheduler.step(noise_pred, t, x).prev_sample
grid = torchvision.utils.make_grid(x, nrow=4)
im = grid.permute(1, 2, 0).cpu().clip(-1, 1) * 0.5 + 0.5
im = Image.fromarray(np.array(im * 255).astype(np.uint8))
# im.save("test.jpeg")
return im
title="""<h1 align="center">Make me a WikiArt</h1>
<p align="center">Try-out of exercise from HF Learn [Difussion Course] </p>
<p align="center">😅 Inference is very very slow 🐌 since I am using HF's free cpu 😉 </p>
<p><center>
<a href="https://huggingface.co/learn/diffusion-course" target="_blank">[HF-Learn]</a>
</center></p>"""
with gr.Blocks() as demo:
gr.HTML(title)
with gr.Row():
with gr.Column():
# Create a radio button with options "no guidance", "color guidance", and "text guidance"
radio = gr.Radio(["no guidance", "color guidance", "text guidance"], label="Choose",value="no guidance")
# Create a textbox that only shows when 'text guidance' is selected
text = gr.Textbox(label="This text only shows when 'text guidance' is selected.", visible=False)
# Create a color picker (not a tuple)
color = gr.ColorPicker(label="color", value="#000000", visible=False)
# Create a slider that shows when any option is selected
slider = gr.Slider(label="guidance_scale", minimum=0, maximum=30, value=3, visible=False)
def update_visibility(radio):
value = radio # Get the selected value from the radio button
if value == "color guidance":
return [gr.Textbox(visible=False),gr.ColorPicker(visible=True),gr.Slider(visible=True)] #make it visible
elif value == "text guidance":
return [gr.Textbox(visible=True),gr.ColorPicker(visible=False),gr.Slider(visible=True)]
else:
return [gr.Textbox(visible=False),gr.ColorPicker(visible=False),gr.Slider(visible=False)]
radio.change(update_visibility, radio,[text,color,slider])
with gr.Column():
outputs = gr.Image(label="result")
with gr.Row():
gen_bttn=gr.Button(value="generate")
gen_bttn.click(generate, inputs=[radio,color,text,slider], outputs=outputs)
demo.queue().launch() |