commit

app.py

@@ -31,7 +31,7 @@ from huggingface_hub import hf_hub_download
 from safetensors.torch import load_file, save_file
 from diffusers import FluxPipeline
 from PIL import Image, ImageDraw, ImageFont
-from transformers import PegasusForConditionalGeneration, PegasusTokenizer
+from transformers import PegasusForConditionalGeneration, PegasusTokenizerFast
 from refiners.fluxion.utils import manual_seed
 from refiners.foundationals.latent_diffusion import Solver, solvers
 from refiners.foundationals.latent_diffusion.stable_diffusion_1.multi_upscaler import (
@@ -502,8 +502,8 @@ image_pipe.enable_model_cpu_offload()
 image_pipe.enable_vae_slicing()
 image_pipe.enable_vae_tiling()
 
-pegasus_tokenizer = PegasusTokenizer.from_pretrained("google/pegasus-xsum")
-pegasus_model = PegasusForConditionalGeneration.from_pretrained("google/pegasus-xsum").to(device)
+pegasus_tokenizer = PegasusTokenizerFast.from_pretrained("google/pegasus-xsum")
+pegasus_model = PegasusForConditionalGeneration.from_pretrained("google/pegasus-xsum")
 
 # functionality
 
@@ -632,7 +632,7 @@ def handle_generation(artist,song,genre,lyrics):
     pos_lyrics = re.sub(f'[{punctuation}]', '', re.sub("([ \t\n]){1,}", " ", lyrics)).lower().strip()
     pos_lyrics_sum = pos_lyrics if pos_lyrics == "" else summarize_text(pos_lyrics)
 
-    neg = f"Sexual, Textual, Labeled, Distorted, Discontinuous, Blurry, Doll-Like, Overly-Plastic, Low-Quality, Painted, Smoothed, Artificial, Phony, Gaudy, Digital Effects."
+    neg = f"Sexuality, Humans, Textual, Labeled, Distorted, Discontinuous, Blurry, Doll-Like, Overly-Plastic, Low-Quality, Painted, Smoothed, Artificial, Phony, Gaudy, Digital Effects."
     q = "\""
     pos = f'Photorealistic, Hyperrealistic, Realistic Photography, High-Quality Photography, Natural, made for the { pos_genre } song "{ pos_song }"{ pos_lyrics_sum if pos_lyrics_sum == "" else ": " + q + pos_lyrics_sum + q }.'