Update instructions

app.py

@@ -106,11 +106,12 @@ duplicate_html = '<a style="display:inline-block" href="https://huggingface.co/s
 html = f"""<h1>LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models</h1>
             <h2>LLM + Stable Diffusion => better prompt understanding in text2image generation 🤩</h2>
             <h2><a href='https://llm-grounded-diffusion.github.io/'>Project Page</a> | <a href='https://bair.berkeley.edu/blog/2023/05/23/lmd/'>5-minute Blog Post</a> | <a href='https://arxiv.org/pdf/2305.13655.pdf'>ArXiv Paper</a> | <a href='https://github.com/TonyLianLong/LLM-groundedDiffusion'>Github</a> | <a href='https://llm-grounded-diffusion.github.io/#citation'>Cite our work</a> if our ideas inspire you.</h2>
-            <p><b>Tips:</b><p>
+            <p><b>Try some examples at the bottom of the page to get started!</b></p>
+            <p><b>Tips:</b></p>
             <p>1. If ChatGPT doesn't generate layout, add/remove the trailing space (added by default) and/or use GPT-4.</p>
             <p>2. You can perform multi-round specification by giving ChatGPT follow-up requests (e.g., make the objects bigger or move the objects).</p>
             <p>3. You can also try prompts in Simplified Chinese. You need to leave "prompt for overall image" empty in this case. If you want to try prompts in another language, translate the first line of last example to your language.</p>
-            <p>4. The diffusion model only runs 20 steps by default in this demo. You can make it run more steps to get higher quality images (or tweak frozen steps/guidance steps for better guidance and coherence).</p>
+            <p>4. The diffusion model only runs {default_num_inference_steps} steps by default in this demo. You can make it run more steps to get higher quality images (or tweak frozen steps/guidance steps for better guidance and coherence).</p>
             <p>5. Duplicate this space and add GPU or clone the space and run locally to skip the queue and run our model faster. (<b>Currently we are using a T4 GPU on this space, which is quite slow, and you can add a A10G to make it 5x faster</b>) {duplicate_html}</p>
             <br/>
             <p>An implementation note (updated): In this demo, we provide a few modes: faster generation by disabling attention/per-box guidance. The standard version describes what is implemented for the paper. You can also set GLIGEN guidance steps ratio to 0 to disable GLIGEN and to see what you get with only the original SD weights.</p>