Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -227,13 +227,9 @@ def uploadNote(prompt,num_inference_steps,guidance_scale,timestamp):
|
|
227 |
f.write(f"Steps: {num_inference_steps} \n")
|
228 |
f.write(f"Guidance Scale: {guidance_scale} \n")
|
229 |
f.write(f"SPACE SETUP: \n")
|
230 |
-
f.write(f"Use Model Dtype: no \n")
|
231 |
-
f.write(f"Model Scheduler: Euler_a all_custom before cuda \n")
|
232 |
-
f.write(f"Model VAE: sdxl-vae to bfloat safetensor=false before cuda then attn_proc / scale factor 8 \n")
|
233 |
f.write(f"Model UNET: ford442/RealVisXL_V5.0_BF16 \n")
|
234 |
upload_to_ftp(filename)
|
235 |
|
236 |
-
|
237 |
def captioning(img):
|
238 |
prompts_array = [
|
239 |
"Adjectives describing this scene are:",
|
@@ -246,15 +242,12 @@ def captioning(img):
|
|
246 |
"The setting of this scene must be located",
|
247 |
# Add more prompts here
|
248 |
]
|
249 |
-
|
250 |
output_prompt=[]
|
251 |
-
|
252 |
# Initial caption generation without a prompt:
|
253 |
inputsa = processor5(images=img, return_tensors="pt").to('cuda')
|
254 |
generated_ids = model5.generate(**inputsa, min_length=42, max_length=42)
|
255 |
generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
256 |
print(generated_text)
|
257 |
-
|
258 |
# Loop through prompts array:
|
259 |
for prompt in prompts_array:
|
260 |
inputs = processor5(images=img, text=prompt, return_tensors="pt").to('cuda')
|
@@ -263,7 +256,6 @@ def captioning(img):
|
|
263 |
response_text = generated_text.replace(prompt, "").strip() #Or could try .split(prompt, 1)[-1].strip()
|
264 |
output_prompt.append(response_text)
|
265 |
print(f"{response_text}\n") # Print only the response text
|
266 |
-
|
267 |
# Continue conversation:
|
268 |
inputf = processor5(images=img, text=generated_text + 'So therefore', return_tensors="pt").to('cuda')
|
269 |
generated_ids = model5.generate(**inputf, max_length=42)
|
@@ -274,7 +266,6 @@ def captioning(img):
|
|
274 |
print(output_prompt)
|
275 |
return output_prompt
|
276 |
|
277 |
-
|
278 |
def expand_prompt(prompt):
|
279 |
system_prompt_rewrite = (
|
280 |
"You are an AI assistant that rewrites image prompts to be more descriptive and detailed."
|
@@ -326,7 +317,7 @@ def expand_prompt(prompt):
|
|
326 |
print('-- filtered prompt 2 --')
|
327 |
print(enhanced_prompt_2)
|
328 |
enh_prompt=[enhanced_prompt,enhanced_prompt_2]
|
329 |
-
|
330 |
|
331 |
@spaces.GPU(duration=40)
|
332 |
def generate_30(
|
|
|
227 |
f.write(f"Steps: {num_inference_steps} \n")
|
228 |
f.write(f"Guidance Scale: {guidance_scale} \n")
|
229 |
f.write(f"SPACE SETUP: \n")
|
|
|
|
|
|
|
230 |
f.write(f"Model UNET: ford442/RealVisXL_V5.0_BF16 \n")
|
231 |
upload_to_ftp(filename)
|
232 |
|
|
|
233 |
def captioning(img):
|
234 |
prompts_array = [
|
235 |
"Adjectives describing this scene are:",
|
|
|
242 |
"The setting of this scene must be located",
|
243 |
# Add more prompts here
|
244 |
]
|
|
|
245 |
output_prompt=[]
|
|
|
246 |
# Initial caption generation without a prompt:
|
247 |
inputsa = processor5(images=img, return_tensors="pt").to('cuda')
|
248 |
generated_ids = model5.generate(**inputsa, min_length=42, max_length=42)
|
249 |
generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
250 |
print(generated_text)
|
|
|
251 |
# Loop through prompts array:
|
252 |
for prompt in prompts_array:
|
253 |
inputs = processor5(images=img, text=prompt, return_tensors="pt").to('cuda')
|
|
|
256 |
response_text = generated_text.replace(prompt, "").strip() #Or could try .split(prompt, 1)[-1].strip()
|
257 |
output_prompt.append(response_text)
|
258 |
print(f"{response_text}\n") # Print only the response text
|
|
|
259 |
# Continue conversation:
|
260 |
inputf = processor5(images=img, text=generated_text + 'So therefore', return_tensors="pt").to('cuda')
|
261 |
generated_ids = model5.generate(**inputf, max_length=42)
|
|
|
266 |
print(output_prompt)
|
267 |
return output_prompt
|
268 |
|
|
|
269 |
def expand_prompt(prompt):
|
270 |
system_prompt_rewrite = (
|
271 |
"You are an AI assistant that rewrites image prompts to be more descriptive and detailed."
|
|
|
317 |
print('-- filtered prompt 2 --')
|
318 |
print(enhanced_prompt_2)
|
319 |
enh_prompt=[enhanced_prompt,enhanced_prompt_2]
|
320 |
+
return enh_prompt
|
321 |
|
322 |
@spaces.GPU(duration=40)
|
323 |
def generate_30(
|