Update app.py

app.py

@@ -85,7 +85,7 @@ model_config.update({
     'class_cond': False,
     'diffusion_steps': 1000,
     'rescale_timesteps': True,
-    'timestep_respacing': '300',  # Modify this value to decrease the number of
+    'timestep_respacing': '90',  # Modify this value to decrease the number of
                                    # timesteps.
     'image_size': 256,
     'learn_sigma': True,
@@ -114,12 +114,12 @@ normalize = transforms.Normalize(mean=[0.48145466, 0.4578275, 0.40821073],
                                  std=[0.26862954, 0.26130258, 0.27577711])
 lpips_model = lpips.LPIPS(net='vgg').to(device)
     
-def inference(text, init_image, skip_timesteps):
+def inference(text, init_image, skip_timesteps, clip_guidance_scale):
     all_frames = []
     prompts = [text]
     image_prompts = []
     batch_size = 1
-    clip_guidance_scale = 1000  # Controls how much the image should look like the prompt.
+    clip_guidance_scale = clip_guidance_scale  # Controls how much the image should look like the prompt.
     tv_scale = 150              # Controls the smoothness of the final output.
     range_scale = 50            # Controls how far out of range RGB values are allowed to be.
     cutn = 16
@@ -208,7 +208,7 @@ def inference(text, init_image, skip_timesteps):
                     all_frames.append(img)
                     tqdm.write(f'Batch {i}, step {j}, output {k}:')
                     #display.display(display.Image(filename))
-    writer = imageio.get_writer('video.mp4', fps=20)
+    writer = imageio.get_writer('video.mp4', fps=5)
     for im in all_frames:
         writer.append_data(np.array(im))
     writer.close()
@@ -217,6 +217,6 @@ def inference(text, init_image, skip_timesteps):
 title = "CLIP Guided Diffusion HQ"
 description = "Gradio demo for CLIP Guided Diffusion. To use it, simply add your text, or click one of the examples to load them. Read more at the links below."
 article = "<p style='text-align: center'> By Katherine Crowson (https://github.com/crowsonkb, https://twitter.com/RiversHaveWings). It uses OpenAI's 256x256 unconditional ImageNet diffusion model (https://github.com/openai/guided-diffusion) together with CLIP (https://github.com/openai/CLIP) to connect text prompts with images. | <a href='https://colab.research.google.com/drive/12a_Wrfi2_gwwAuN3VvMTwVMz9TfqctNj' target='_blank'>Colab</a></p>"
-iface = gr.Interface(inference, inputs=["text",gr.inputs.Image(type="file", label='initial image (optional)', optional=True),gr.inputs.Slider(minimum=0, maximum=150, step=1, default=0, label="skip_timesteps")], outputs=["image","video"], title=title, description=description, article=article, examples=[["coral reef city by artistation artists"]],
+iface = gr.Interface(inference, inputs=["text",gr.inputs.Image(type="file", label='initial image (optional)', optional=True),gr.inputs.Slider(minimum=0, maximum=45, step=1, default=0, label="skip_timesteps"), gr.inputs.Slider(minimum=0, maximum=1000, step=1, default=700, label="clip guidance scale (Controls how much the image should look like the prompt.)")], outputs=["image","video"], title=title, description=description, article=article, examples=[["coral reef city by artistation artists"]],
     enable_queue=True)
 iface.launch()