update

README.md

@@ -37,6 +37,8 @@ accelerate launch \
     --report_to="tensorboard" \
     --logging_dir="./nyc-ad-model/logs" 
 
+fine tune a trained model: --pretrained_model_name_or_path="./nyc-ad-model/checkpoint-400" \
+
 
 
 export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True

image_download.py

@@ -11,7 +11,7 @@ flickr = flickrapi.FlickrAPI(FLICKR_PUBLIC, FLICKR_SECRET, format='parsed-json')
 
 # Search for images with relevant tags
 results = flickr.photos.search(
-    text='brooklyn advertisement',
+    text='advertisement',
     per_page=50,
     media='photos',
     sort='relevance',

inference.py

@@ -7,8 +7,8 @@ pipe = StableDiffusionPipeline.from_pretrained(
     torch_dtype=torch.float16,
 ).to("cuda")  # use "cpu" if no GPU
 
-prompt = "fried chicken advertisement poster: a fried chicken in brooklyn street"
-image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
+prompt = "brand name: xyc, fried chicken advertisement poster: a fried chicken in brooklyn street"
+image = pipe(prompt, num_inference_steps=500, guidance_scale=7.5).images[0]
 
 # Display or save the image
 image.save("output_nyc_ad.png")