Update app.py

app.py

@@ -24,7 +24,7 @@ from composable_diffusion.model_creation import create_model_and_diffusion as cr
 from composable_diffusion.model_creation import model_and_diffusion_defaults as model_and_diffusion_defaults_for_clevr
 
 
-from PIL import Image
+from PIL import Image, ImageDraw, ImageFont
 
 from torch import autocast
 from diffusers import StableDiffusionPipeline
@@ -316,7 +316,20 @@ def compose(prompt, version, guidance_scale, steps):
             model.to(cpu)
             model_up.to(cpu)
             clevr_model.to(device)
-            return compose_clevr_objects(prompt, guidance_scale, steps)
+            # simple check
+            is_text = True
+            for char in prompt:
+                if char.isdigit():
+                    is_text = False
+                    break
+            if is_text:
+                img = Image.new('RGB', (512, 512), color=(255, 255, 255))
+                d = ImageDraw.Draw(img)
+                font = ImageFont.load_default()
+                d.text((0, 256), "input should be similar to the example using 2D coordinates.", fill=(0, 0, 0), font=font)
+                return img
+            else:
+                return compose_clevr_objects(prompt, guidance_scale, steps)
 
 
 examples_1 = 'a camel | a forest'
@@ -339,7 +352,7 @@ examples = [
 import gradio as gr
 
 title = 'Compositional Visual Generation with Composable Diffusion Models'
-description = '<p>Demo for Composable Diffusion<ul><li>~30s per GLIDE/Stable-Diffusion example</li><li>~10s per CLEVR Object example</li>(<b>Note</b>: time is varied depending on what gpu is used.)</ul></p><p>See more information from our <a href="https://energy-based-model.github.io/Compositional-Visual-Generation-with-Composable-Diffusion-Models/">Project Page</a>.</p><ul><li>One version is based on the released <a href="https://github.com/openai/glide-text2im">GLIDE</a> and <a href="https://github.com/CompVis/stable-diffusion/">Stable Diffusion</a> for composing natural language description.</li><li>Another is based on our pre-trained CLEVR Object Model for composing objects. <br>(<b>Note</b>: We recommend using <b><i>x</i></b> in range <b><i>[0.1, 0.9]</i></b> and <b><i>y</i></b> in range <b><i>[0.25, 0.7]</i></b>, since the training dataset labels are in given ranges.)</li></ul><p>When composing  multiple sentences, use `|` as the delimiter, see given examples below.</p><p><b>Note</b>: When using more steps, the results can improve.</p>'
+description = '<p>Demo for Composable Diffusion<ul><li>~30s per GLIDE/Stable-Diffusion example</li><li>~10s per CLEVR Object example</li>(<b>Note</b>: time is varied depending on what gpu is used.)</ul></p><p>See more information from our <a href="https://energy-based-model.github.io/Compositional-Visual-Generation-with-Composable-Diffusion-Models/">Project Page</a>.</p><ul><li>One version is based on the released <a href="https://github.com/openai/glide-text2im">GLIDE</a> and <a href="https://github.com/CompVis/stable-diffusion/">Stable Diffusion</a> for composing natural language description.</li><li>Another is based on our pre-trained CLEVR Object Model for composing objects. <br>(<b>Note</b>: We recommend using <b><i>x</i></b> in range <b><i>[0.1, 0.9]</i></b> and <b><i>y</i></b> in range <b><i>[0.25, 0.7]</i></b>, since the training dataset labels are in given ranges.)</li></ul><p>When composing  multiple sentences, use `|` as the delimiter, see given examples below.</p><p><b>Note</b>: When using Stable Diffusion, black images will be returned if the given prompt is detected as problematic.</p>'
 
 iface = gr.Interface(compose,
                      inputs=[