Spaces:
Running
on
Zero
Running
on
Zero
hav4ik
commited on
Commit
•
2338872
1
Parent(s):
4d8d3ff
app
Browse files- app.py +577 -117
- requirements.txt +7 -1
- style.css +16 -0
app.py
CHANGED
@@ -1,142 +1,602 @@
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
import numpy as np
|
3 |
-
import
|
4 |
-
|
5 |
-
from diffusers import DiffusionPipeline
|
6 |
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
9 |
-
model_repo_id = "stabilityai/sdxl-turbo" #Replace to the model you would like to use
|
10 |
|
11 |
if torch.cuda.is_available():
|
12 |
-
|
|
|
|
|
|
|
13 |
else:
|
14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
-
pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
|
17 |
-
pipe = pipe.to(device)
|
18 |
|
19 |
-
MAX_SEED = np.iinfo(np.int32).max
|
20 |
-
MAX_IMAGE_SIZE = 1024
|
21 |
|
22 |
-
|
23 |
-
def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
|
24 |
|
|
|
25 |
if randomize_seed:
|
26 |
seed = random.randint(0, MAX_SEED)
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
""
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
)
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
value=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
)
|
112 |
-
|
113 |
-
with gr.Row():
|
114 |
-
|
115 |
guidance_scale = gr.Slider(
|
116 |
label="Guidance scale",
|
117 |
-
minimum=0.
|
118 |
maximum=10.0,
|
119 |
step=0.1,
|
120 |
-
value=
|
121 |
)
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
127 |
step=1,
|
128 |
-
value=
|
129 |
)
|
130 |
-
|
131 |
-
gr.
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
140 |
)
|
141 |
|
142 |
-
|
|
|
|
1 |
+
import os
|
2 |
+
import random
|
3 |
+
|
4 |
import gradio as gr
|
5 |
import numpy as np
|
6 |
+
import PIL.Image
|
7 |
+
from PIL import ImageOps
|
|
|
8 |
import torch
|
9 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
10 |
+
from transformers import BitsAndBytesConfig
|
11 |
+
import torchvision.transforms.functional as TF
|
12 |
+
from diffusers import (
|
13 |
+
AutoencoderKL,
|
14 |
+
EulerAncestralDiscreteScheduler,
|
15 |
+
StableDiffusionXLAdapterPipeline,
|
16 |
+
T2IAdapter,
|
17 |
+
)
|
18 |
+
|
19 |
+
import urllib.parse
|
20 |
+
import requests
|
21 |
+
from io import BytesIO
|
22 |
+
import json
|
23 |
+
|
24 |
+
from pathlib import Path
|
25 |
+
import uuid
|
26 |
+
import os, uuid
|
27 |
+
from azure.identity import DefaultAzureCredential
|
28 |
+
from azure.storage.blob import BlobServiceClient, BlobClient, ContainerClient
|
29 |
+
|
30 |
+
from datetime import datetime
|
31 |
+
|
32 |
+
|
33 |
+
|
34 |
+
class DEFAULTS:
|
35 |
+
NEGATIVE_PROMPT = " extra digit, fewer digits, cropped, worst quality, low quality, glitch, deformed, mutated, ugly, disfigured"
|
36 |
+
REWRITING_PROMPT = (
|
37 |
+
"Rewrite the image caption by making it shorter (but retain all information about relative position), "
|
38 |
+
"remove information about style of objects or colors of background and foreground, and, most importantly, remove all details "
|
39 |
+
"that suggests it is a sketch. Write it as a Google image search query:"
|
40 |
+
)
|
41 |
+
MOONDREAM_PROMPT = "Describe this image."
|
42 |
+
NUM_STEPS = 25
|
43 |
+
GUIDANCE_SCALE = 5
|
44 |
+
ADAPTER_CONDITIONING_SCALE = 0.8
|
45 |
+
ADAPTER_CONDITIONING_FACTOR = 0.8
|
46 |
+
SEED = 1231245
|
47 |
+
RANDOMIZE_SEED = True
|
48 |
+
|
49 |
+
|
50 |
+
|
51 |
+
DESCRIPTION = '''# Sketch to Image/Caption to Bing Search :)
|
52 |
+
'''
|
53 |
+
|
54 |
+
if not torch.cuda.is_available():
|
55 |
+
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
56 |
+
|
57 |
+
style_list = [
|
58 |
+
{
|
59 |
+
"name": "(No style)",
|
60 |
+
"prompt": "{prompt}",
|
61 |
+
"negative_prompt": "",
|
62 |
+
},
|
63 |
+
{
|
64 |
+
"name": "Cinematic",
|
65 |
+
"prompt": "cinematic still {prompt} . emotional, harmonious, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy",
|
66 |
+
"negative_prompt": "anime, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured",
|
67 |
+
},
|
68 |
+
{
|
69 |
+
"name": "3D Model",
|
70 |
+
"prompt": "professional 3d model {prompt} . octane render, highly detailed, volumetric, dramatic lighting",
|
71 |
+
"negative_prompt": "ugly, deformed, noisy, low poly, blurry, painting",
|
72 |
+
},
|
73 |
+
{
|
74 |
+
"name": "Anime",
|
75 |
+
"prompt": "anime artwork {prompt} . anime style, key visual, vibrant, studio anime, highly detailed",
|
76 |
+
"negative_prompt": "photo, deformed, black and white, realism, disfigured, low contrast",
|
77 |
+
},
|
78 |
+
{
|
79 |
+
"name": "Digital Art",
|
80 |
+
"prompt": "concept art {prompt} . digital artwork, illustrative, painterly, matte painting, highly detailed",
|
81 |
+
"negative_prompt": "photo, photorealistic, realism, ugly",
|
82 |
+
},
|
83 |
+
{
|
84 |
+
"name": "Photographic",
|
85 |
+
"prompt": "cinematic photo {prompt} . 35mm photograph, film, bokeh, professional, 4k, highly detailed",
|
86 |
+
"negative_prompt": "drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly",
|
87 |
+
},
|
88 |
+
{
|
89 |
+
"name": "Pixel art",
|
90 |
+
"prompt": "pixel-art {prompt} . low-res, blocky, pixel art style, 8-bit graphics",
|
91 |
+
"negative_prompt": "sloppy, messy, blurry, noisy, highly detailed, ultra textured, photo, realistic",
|
92 |
+
},
|
93 |
+
{
|
94 |
+
"name": "Fantasy art",
|
95 |
+
"prompt": "ethereal fantasy concept art of {prompt} . magnificent, celestial, ethereal, painterly, epic, majestic, magical, fantasy art, cover art, dreamy",
|
96 |
+
"negative_prompt": "photographic, realistic, realism, 35mm film, dslr, cropped, frame, text, deformed, glitch, noise, noisy, off-center, deformed, cross-eyed, closed eyes, bad anatomy, ugly, disfigured, sloppy, duplicate, mutated, black and white",
|
97 |
+
},
|
98 |
+
{
|
99 |
+
"name": "Neonpunk",
|
100 |
+
"prompt": "neonpunk style {prompt} . cyberpunk, vaporwave, neon, vibes, vibrant, stunningly beautiful, crisp, detailed, sleek, ultramodern, magenta highlights, dark purple shadows, high contrast, cinematic, ultra detailed, intricate, professional",
|
101 |
+
"negative_prompt": "painting, drawing, illustration, glitch, deformed, mutated, cross-eyed, ugly, disfigured",
|
102 |
+
},
|
103 |
+
{
|
104 |
+
"name": "Manga",
|
105 |
+
"prompt": "manga style {prompt} . vibrant, high-energy, detailed, iconic, Japanese comic style",
|
106 |
+
"negative_prompt": "ugly, deformed, noisy, blurry, low contrast, realism, photorealistic, Western comic style",
|
107 |
+
},
|
108 |
+
]
|
109 |
+
|
110 |
+
styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}
|
111 |
+
STYLE_NAMES = list(styles.keys())
|
112 |
+
DEFAULT_STYLE_NAME = "Photographic" # "(No style)"
|
113 |
+
|
114 |
+
|
115 |
+
def apply_style(style_name: str, positive: str, negative: str = "") -> tuple[str, str]:
|
116 |
+
p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
|
117 |
+
return p.replace("{prompt}", positive), n + negative
|
118 |
+
|
119 |
+
|
120 |
+
# with open("azure_connection_string.txt", "r") as f:
|
121 |
+
# CONNECTION_STRING = f.read().strip()
|
122 |
+
CONNECTION_STRING = os.getenv("AZURE_CONNECTION_STRING")
|
123 |
+
|
124 |
+
|
125 |
+
def upload_pil_image_to_azure(image, connection_string=CONNECTION_STRING):
|
126 |
+
image_name = f"{uuid.uuid4()}.png"
|
127 |
+
image_bytes = BytesIO()
|
128 |
+
image.save(image_bytes, format="PNG")
|
129 |
+
image_bytes.seek(0)
|
130 |
+
|
131 |
+
try:
|
132 |
+
# Create the BlobServiceClient object
|
133 |
+
blob_service_client = BlobServiceClient.from_connection_string(connection_string)
|
134 |
+
# Create a blob client using the local file name as the name for the blob
|
135 |
+
blob_client = blob_service_client.get_blob_client(container="blob-image-hosting", blob=image_name)
|
136 |
+
# Upload the created file and retrieve the URL
|
137 |
+
blob_client.upload_blob(image_bytes)
|
138 |
+
file_url = blob_client.url
|
139 |
+
except Exception as ex:
|
140 |
+
print('Exception:')
|
141 |
+
print(ex)
|
142 |
+
file_url = None
|
143 |
+
# If this function did not fail, upload was successful
|
144 |
+
return file_url
|
145 |
|
|
|
|
|
146 |
|
147 |
if torch.cuda.is_available():
|
148 |
+
if torch.cuda.device_count() > 1:
|
149 |
+
device_0, device_1 = torch.device("cuda:0"), torch.device("cuda:1")
|
150 |
+
else:
|
151 |
+
device_0, device_1 = torch.device("cuda:0"), torch.device("cuda:0")
|
152 |
else:
|
153 |
+
device_0, device_1 = torch.device("cpu"), torch.device("cpu")
|
154 |
+
|
155 |
+
if torch.cuda.is_available():
|
156 |
+
model_id = "stabilityai/stable-diffusion-xl-base-1.0"
|
157 |
+
adapter = T2IAdapter.from_pretrained(
|
158 |
+
"TencentARC/t2i-adapter-sketch-sdxl-1.0", torch_dtype=torch.float16, variant="fp16"
|
159 |
+
)
|
160 |
+
scheduler = EulerAncestralDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
|
161 |
+
pipe = StableDiffusionXLAdapterPipeline.from_pretrained(
|
162 |
+
model_id,
|
163 |
+
vae=AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16),
|
164 |
+
adapter=adapter,
|
165 |
+
scheduler=scheduler,
|
166 |
+
torch_dtype=torch.float16,
|
167 |
+
variant="fp16",
|
168 |
+
)
|
169 |
+
pipe.to(device_0)
|
170 |
+
else:
|
171 |
+
pipe = None
|
172 |
|
|
|
|
|
173 |
|
|
|
|
|
174 |
|
175 |
+
MAX_SEED = np.iinfo(np.int32).max
|
|
|
176 |
|
177 |
+
def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
|
178 |
if randomize_seed:
|
179 |
seed = random.randint(0, MAX_SEED)
|
180 |
+
return seed
|
181 |
+
|
182 |
+
nf4_config = BitsAndBytesConfig(
|
183 |
+
load_in_4bit=True,
|
184 |
+
bnb_4bit_quant_type="nf4",
|
185 |
+
bnb_4bit_use_double_quant=True,
|
186 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
187 |
+
)
|
188 |
+
|
189 |
+
vlmodel_id = "vikhyatk/moondream2"
|
190 |
+
vlmodel_revision = "2024-07-23"
|
191 |
+
vlmodel = AutoModelForCausalLM.from_pretrained(
|
192 |
+
vlmodel_id, trust_remote_code=True, revision=vlmodel_revision, device_map={"": device_1},
|
193 |
+
torch_dtype=torch.float16, attn_implementation="flash_attention_2")
|
194 |
+
vltokenizer = AutoTokenizer.from_pretrained(vlmodel_id, revision=vlmodel_revision)
|
195 |
+
|
196 |
+
rewrite_model_name = "meta-llama/Meta-Llama-3.1-8B-Instruct"
|
197 |
+
rewrite_model = AutoModelForCausalLM.from_pretrained(
|
198 |
+
rewrite_model_name,
|
199 |
+
device_map={"": device_1},
|
200 |
+
quantization_config=nf4_config,
|
201 |
+
# load_in_8bit=True,
|
202 |
+
torch_dtype=torch.bfloat16,
|
203 |
+
attn_implementation="flash_attention_2")
|
204 |
+
rewrite_tokenizer = AutoTokenizer.from_pretrained(rewrite_model_name)
|
205 |
+
|
206 |
+
|
207 |
+
def caption_image_with_recaption(pil_image, moondream_prompt, rewriting_prompt, user_prompt=""):
|
208 |
+
enc_image = vlmodel.encode_image(pil_image)
|
209 |
+
img_caption = vlmodel.answer_question(enc_image, moondream_prompt, vltokenizer)
|
210 |
+
rewritten_caption = rewrite_prompt(img_caption, rewriting_prompt, user_prompt=user_prompt)
|
211 |
+
rewritten_caption = rewritten_caption.strip('"').replace("\n", " ")
|
212 |
+
return img_caption, rewritten_caption
|
213 |
+
|
214 |
+
|
215 |
+
def rewrite_prompt(image_cap: str, guide: str, user_prompt: str = "") -> str:
|
216 |
+
prompt = f"{guide}\n{image_cap}"
|
217 |
+
messages = [
|
218 |
+
{"role": "system", "content": "You are a helpful assistant."},
|
219 |
+
{"role": "user", "content": prompt}
|
220 |
+
]
|
221 |
+
text = rewrite_tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
222 |
+
model_inputs = rewrite_tokenizer([text], return_tensors="pt").to(device_1)
|
223 |
+
generated_ids = rewrite_model.generate(model_inputs.input_ids, max_new_tokens=128)
|
224 |
+
generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
|
225 |
+
response = rewrite_tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
226 |
+
return response
|
227 |
+
|
228 |
+
|
229 |
+
def run_full(
|
230 |
+
image,
|
231 |
+
user_prompt: str,
|
232 |
+
negative_prompt: str,
|
233 |
+
rewriting_prompt: str,
|
234 |
+
moondream_prompt: str,
|
235 |
+
style_name: str = DEFAULT_STYLE_NAME,
|
236 |
+
num_steps: int = 25,
|
237 |
+
guidance_scale: float = 5,
|
238 |
+
adapter_conditioning_scale: float = 0.8,
|
239 |
+
adapter_conditioning_factor: float = 0.8,
|
240 |
+
seed: int = 0,
|
241 |
+
progress=None,
|
242 |
+
) -> PIL.Image.Image:
|
243 |
+
# image is a white background with black sketch
|
244 |
+
image = ImageOps.invert(image)
|
245 |
+
# Threshold the image to get a binary sketch
|
246 |
+
image = TF.to_tensor(image) > 0.5
|
247 |
+
image = TF.to_pil_image(image.to(torch.float32))
|
248 |
+
|
249 |
+
full_log = []
|
250 |
+
if user_prompt == "":
|
251 |
+
pre_caption = True
|
252 |
+
start_time = datetime.now()
|
253 |
+
img_caption, rewritten_caption = caption_image_with_recaption(
|
254 |
+
pil_image=image, rewriting_prompt=rewriting_prompt, moondream_prompt=moondream_prompt)
|
255 |
+
full_log.append(f"Combined captioning time: {datetime.now() - start_time}")
|
256 |
+
full_log.append(f"img_caption (pre): {img_caption}")
|
257 |
+
full_log.append(f"rewritten_caption (pre): {rewritten_caption}")
|
258 |
+
drawing_prompt = rewritten_caption
|
259 |
+
else:
|
260 |
+
pre_caption = False
|
261 |
+
drawing_prompt = user_prompt
|
262 |
+
full_log.append(f"Pre-caption: {pre_caption}")
|
263 |
+
|
264 |
+
# Generate image
|
265 |
+
start_time = datetime.now()
|
266 |
+
drawing_prompt, negative_prompt = apply_style(style_name, drawing_prompt, negative_prompt)
|
267 |
+
generator = torch.Generator(device=device_0).manual_seed(seed)
|
268 |
+
out_img = pipe(
|
269 |
+
prompt=drawing_prompt,
|
270 |
+
negative_prompt=negative_prompt,
|
271 |
+
image=image,
|
272 |
+
num_inference_steps=num_steps,
|
273 |
+
generator=generator,
|
274 |
+
guidance_scale=guidance_scale,
|
275 |
+
adapter_conditioning_scale=adapter_conditioning_scale,
|
276 |
+
adapter_conditioning_factor=adapter_conditioning_factor,
|
277 |
+
).images[0]
|
278 |
+
full_log.append(f"Image generation time: {datetime.now() - start_time}")
|
279 |
+
|
280 |
+
if not pre_caption:
|
281 |
+
start_time = datetime.now()
|
282 |
+
img_caption, rewritten_caption = caption_image_with_recaption(
|
283 |
+
pil_image=out_img,
|
284 |
+
rewriting_prompt=rewriting_prompt,
|
285 |
+
moondream_prompt=moondream_prompt,
|
286 |
+
user_prompt=user_prompt)
|
287 |
+
full_log.append(f"Combined captioning time: {datetime.now() - start_time}")
|
288 |
+
full_log.append(f"img_caption (post): {img_caption}")
|
289 |
+
full_log.append(f"rewritten_caption (post): {rewritten_caption}")
|
290 |
|
291 |
+
# SERP query
|
292 |
+
bing_serp_query = f"https://www.bing.com/images/search?q={urllib.parse.quote(rewritten_caption)}"
|
293 |
+
md_text = f"### Bing search query\n[{bing_serp_query}]({bing_serp_query})\n"
|
294 |
+
|
295 |
+
# Visual Search query
|
296 |
+
out_img_imgur_url = upload_pil_image_to_azure(out_img)
|
297 |
+
if out_img_imgur_url is None:
|
298 |
+
md_text += "### Bing Visual Search\n**Error:** Failed to upload image to Azure Blob Storage\n"
|
299 |
+
bing_image_search_url = "https://www.bing.com/images"
|
300 |
+
else:
|
301 |
+
imgur_url_quote = urllib.parse.quote(out_img_imgur_url)
|
302 |
+
bing_image_search_url = f"https://www.bing.com/images/search?view=detailv2&iss=SBI&form=SBIIRP&q=imgurl:{imgur_url_quote}"
|
303 |
+
md_text += f"### Bing Visual Search\n[{bing_image_search_url}]({bing_image_search_url})\n"
|
304 |
+
|
305 |
+
# Debug info
|
306 |
+
md_text += f"### Debug: sketch caption\n{img_caption}\n\n### Debug: rewritten caption\n{rewritten_caption}\n"
|
307 |
+
|
308 |
+
# Full log dump
|
309 |
+
md_text += f"### Debug: full log\n{'<br>'.join(full_log)}"
|
310 |
+
|
311 |
+
# return dict
|
312 |
+
return {
|
313 |
+
"image": out_img,
|
314 |
+
"text_search_url": bing_serp_query,
|
315 |
+
"visual_search_url": bing_image_search_url,
|
316 |
+
"logs": md_text,
|
317 |
+
}
|
318 |
+
|
319 |
+
|
320 |
+
def run_full_gradio(
|
321 |
+
image,
|
322 |
+
user_prompt: str,
|
323 |
+
negative_prompt: str,
|
324 |
+
rewriting_prompt: str,
|
325 |
+
moondream_prompt: str,
|
326 |
+
style_name: str = DEFAULT_STYLE_NAME,
|
327 |
+
num_steps: int = 25,
|
328 |
+
guidance_scale: float = 5,
|
329 |
+
adapter_conditioning_scale: float = 0.8,
|
330 |
+
adapter_conditioning_factor: float = 0.8,
|
331 |
+
seed: int = 0,
|
332 |
+
progress=gr.Progress(track_tqdm=True),
|
333 |
+
) -> PIL.Image.Image:
|
334 |
+
image = image['composite']
|
335 |
+
background = PIL.Image.new('RGBA', image.size, (255, 255, 255))
|
336 |
+
alpha_composite = PIL.Image.alpha_composite(background, image)
|
337 |
+
image = alpha_composite.convert("RGB")
|
338 |
+
|
339 |
+
results = run_full(
|
340 |
+
image=image,
|
341 |
+
user_prompt=user_prompt,
|
342 |
+
negative_prompt=negative_prompt,
|
343 |
+
rewriting_prompt=rewriting_prompt,
|
344 |
+
moondream_prompt=moondream_prompt,
|
345 |
+
style_name=style_name,
|
346 |
+
num_steps=num_steps,
|
347 |
+
guidance_scale=guidance_scale,
|
348 |
+
adapter_conditioning_scale=adapter_conditioning_scale,
|
349 |
+
adapter_conditioning_factor=adapter_conditioning_factor,
|
350 |
+
seed=seed,
|
351 |
+
progress=progress,
|
352 |
+
)
|
353 |
+
|
354 |
+
# construct markdown output
|
355 |
+
return results["image"], results["logs"]
|
356 |
+
|
357 |
+
|
358 |
+
def run_full_api(
|
359 |
+
image_url: str,
|
360 |
+
user_prompt: str,
|
361 |
+
progress=gr.Progress(track_tqdm=True),
|
362 |
+
) -> str:
|
363 |
+
seed = randomize_seed_fn(0, True)
|
364 |
+
image = PIL.Image.open(BytesIO(requests.get(image_url).content))
|
365 |
+
results = run_full(
|
366 |
+
image=image, user_prompt=user_prompt,
|
367 |
+
negative_prompt=DEFAULTS.NEGATIVE_PROMPT,
|
368 |
+
rewriting_prompt=DEFAULTS.REWRITING_PROMPT,
|
369 |
+
moondream_prompt=DEFAULTS.MOONDREAM_PROMPT,
|
370 |
+
style_name=DEFAULT_STYLE_NAME,
|
371 |
+
num_steps=DEFAULTS.NUM_STEPS,
|
372 |
+
guidance_scale=DEFAULTS.GUIDANCE_SCALE,
|
373 |
+
adapter_conditioning_scale=DEFAULTS.ADAPTER_CONDITIONING_SCALE,
|
374 |
+
adapter_conditioning_factor=DEFAULTS.ADAPTER_CONDITIONING_FACTOR,
|
375 |
+
seed=seed)
|
376 |
+
return results["text_search_url"], results["visual_search_url"], results["logs"]
|
377 |
+
|
378 |
+
|
379 |
+
def run_caponly(
|
380 |
+
image,
|
381 |
+
rewriting_prompt: str,
|
382 |
+
moondream_prompt: str,
|
383 |
+
seed: int = 0,
|
384 |
+
progress=None,
|
385 |
+
) -> PIL.Image.Image:
|
386 |
+
# image is a white background with black sketch
|
387 |
+
image = ImageOps.invert(image)
|
388 |
+
# Threshold the image to get a binary sketch
|
389 |
+
image = TF.to_tensor(image) > 0.5
|
390 |
+
image = TF.to_pil_image(image.to(torch.float32))
|
391 |
+
|
392 |
+
full_log = []
|
393 |
+
start_time = datetime.now()
|
394 |
+
img_caption, rewritten_caption = caption_image_with_recaption(
|
395 |
+
pil_image=image, rewriting_prompt=rewriting_prompt, moondream_prompt=moondream_prompt)
|
396 |
+
full_log.append(f"Combined captioning time: {datetime.now() - start_time}")
|
397 |
+
full_log.append(f"img_caption (pre): {img_caption}")
|
398 |
+
full_log.append(f"rewritten_caption (pre): {rewritten_caption}")
|
399 |
+
final_prompt = rewritten_caption
|
400 |
+
|
401 |
+
# SERP query
|
402 |
+
bing_serp_query = f"https://www.bing.com/images/search?q={urllib.parse.quote(rewritten_caption)}"
|
403 |
+
md_text = f"### Bing search query\n[{bing_serp_query}]({bing_serp_query})\n"
|
404 |
+
|
405 |
+
# Debug info
|
406 |
+
md_text += f"### Debug: sketch caption\n{img_caption}\n\n### Debug: rewritten caption\n{rewritten_caption}\n"
|
407 |
+
|
408 |
+
# Full log dump
|
409 |
+
md_text += f"### Debug: full log\n{'<br>'.join(full_log)}"
|
410 |
+
|
411 |
+
# return dict
|
412 |
+
return {
|
413 |
+
"text_search_url": bing_serp_query,
|
414 |
+
"logs": md_text,
|
415 |
+
}
|
416 |
+
|
417 |
+
|
418 |
+
def run_caponly_api(
|
419 |
+
image_url: str,
|
420 |
+
progress=gr.Progress(track_tqdm=True),
|
421 |
+
) -> str:
|
422 |
+
seed = randomize_seed_fn(0, True)
|
423 |
+
image = PIL.Image.open(BytesIO(requests.get(image_url).content))
|
424 |
+
results = run_caponly(
|
425 |
+
image=image,
|
426 |
+
rewriting_prompt=DEFAULTS.REWRITING_PROMPT,
|
427 |
+
moondream_prompt=DEFAULTS.MOONDREAM_PROMPT,
|
428 |
+
seed=seed)
|
429 |
+
return results["text_search_url"], results["logs"]
|
430 |
+
|
431 |
+
|
432 |
+
with gr.Blocks(css="style.css") as demo:
|
433 |
+
gr.Markdown(DESCRIPTION, elem_id="description")
|
434 |
+
gr.DuplicateButton(
|
435 |
+
value="Duplicate Space for private use",
|
436 |
+
elem_id="duplicate-button",
|
437 |
+
visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
|
438 |
+
)
|
439 |
+
|
440 |
+
with gr.Row():
|
441 |
+
with gr.Column():
|
442 |
+
with gr.Group():
|
443 |
+
image = gr.Sketchpad(
|
444 |
+
# sources=["canvas"],
|
445 |
+
# tool="sketch",
|
446 |
+
type="pil",
|
447 |
+
image_mode="RGBA",
|
448 |
+
# invert_colors=True,
|
449 |
+
layers=False,
|
450 |
+
canvas_size=(1024, 1024),
|
451 |
+
brush=gr.Brush(
|
452 |
+
default_color="black",
|
453 |
+
colors=None,
|
454 |
+
default_size=4,
|
455 |
+
color_mode="fixed",
|
456 |
+
),
|
457 |
+
eraser=gr.Eraser(),
|
458 |
+
height=440,
|
459 |
+
)
|
460 |
+
prompt = gr.Textbox(label="Prompt")
|
461 |
+
style = gr.Dropdown(label="Style", choices=STYLE_NAMES, value=DEFAULT_STYLE_NAME)
|
462 |
+
run_button = gr.Button("Run")
|
463 |
+
with gr.Accordion("Advanced options", open=False):
|
464 |
+
negative_prompt = gr.Textbox(
|
465 |
+
label="Negative prompt",
|
466 |
+
value=DEFAULTS.NEGATIVE_PROMPT,
|
467 |
)
|
468 |
+
rewriting_prompt = gr.Textbox(
|
469 |
+
label="Rewriting prompt",
|
470 |
+
value=DEFAULTS.REWRITING_PROMPT,
|
471 |
+
)
|
472 |
+
moondream_prompt = gr.Textbox(
|
473 |
+
label="Moondream prompt",
|
474 |
+
value=DEFAULTS.MOONDREAM_PROMPT,
|
475 |
+
)
|
476 |
+
num_steps = gr.Slider(
|
477 |
+
label="Number of steps",
|
478 |
+
minimum=1,
|
479 |
+
maximum=50,
|
480 |
+
step=1,
|
481 |
+
value=DEFAULTS.NUM_STEPS,
|
482 |
)
|
|
|
|
|
|
|
483 |
guidance_scale = gr.Slider(
|
484 |
label="Guidance scale",
|
485 |
+
minimum=0.1,
|
486 |
maximum=10.0,
|
487 |
step=0.1,
|
488 |
+
value=DEFAULTS.GUIDANCE_SCALE,
|
489 |
)
|
490 |
+
adapter_conditioning_scale = gr.Slider(
|
491 |
+
label="Adapter conditioning scale",
|
492 |
+
minimum=0.5,
|
493 |
+
maximum=1,
|
494 |
+
step=0.1,
|
495 |
+
value=DEFAULTS.ADAPTER_CONDITIONING_SCALE,
|
496 |
+
)
|
497 |
+
adapter_conditioning_factor = gr.Slider(
|
498 |
+
label="Adapter conditioning factor",
|
499 |
+
info="Fraction of timesteps for which adapter should be applied",
|
500 |
+
minimum=0.5,
|
501 |
+
maximum=1,
|
502 |
+
step=0.1,
|
503 |
+
value=DEFAULTS.ADAPTER_CONDITIONING_FACTOR,
|
504 |
+
)
|
505 |
+
seed = gr.Slider(
|
506 |
+
label="Seed",
|
507 |
+
minimum=0,
|
508 |
+
maximum=MAX_SEED,
|
509 |
step=1,
|
510 |
+
value=0,
|
511 |
)
|
512 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
513 |
+
with gr.Column():
|
514 |
+
result_img = gr.Image(label="Result", height=400, interactive=False)
|
515 |
+
result_caption = gr.Markdown(label="Image caption")
|
516 |
+
result = [result_img, result_caption]
|
517 |
+
|
518 |
+
with gr.Row():
|
519 |
+
gr.Markdown("# API endpoints", elem_id="description")
|
520 |
+
with gr.Row():
|
521 |
+
with gr.Column():
|
522 |
+
with gr.Accordion("Full Experience API", open=False):
|
523 |
+
api_fullexp_image_url = gr.Textbox(label="Image URL")
|
524 |
+
api_fullexp_user_prompt = gr.Textbox(label="User prompt")
|
525 |
+
api_fullexp_run_button = gr.Button("Run API")
|
526 |
+
api_fullexp_text_search_url = gr.Textbox(label="Text search URL")
|
527 |
+
api_fullexp_visual_search_url = gr.Textbox(label="Visual search URL")
|
528 |
+
api_fullexp_logs = gr.Markdown(label="Logs")
|
529 |
+
with gr.Column():
|
530 |
+
with gr.Accordion("Caption Only API", open=False):
|
531 |
+
api_caponly_image_url = gr.Textbox(label="Image URL")
|
532 |
+
api_caponly_run_button = gr.Button("Run API")
|
533 |
+
api_caponly_text_search_url = gr.Textbox(label="Text search URL")
|
534 |
+
api_caponly_logs = gr.Markdown(label="Logs")
|
535 |
+
|
536 |
+
# Gradio components interconnections
|
537 |
+
inputs = [
|
538 |
+
image,
|
539 |
+
prompt,
|
540 |
+
negative_prompt,
|
541 |
+
rewriting_prompt,
|
542 |
+
moondream_prompt,
|
543 |
+
style,
|
544 |
+
num_steps,
|
545 |
+
guidance_scale,
|
546 |
+
adapter_conditioning_scale,
|
547 |
+
adapter_conditioning_factor,
|
548 |
+
seed,
|
549 |
+
]
|
550 |
+
prompt.submit(
|
551 |
+
fn=randomize_seed_fn,
|
552 |
+
inputs=[seed, randomize_seed],
|
553 |
+
outputs=seed,
|
554 |
+
queue=False,
|
555 |
+
api_name=False,
|
556 |
+
).then(
|
557 |
+
fn=run_full_gradio,
|
558 |
+
inputs=inputs,
|
559 |
+
outputs=result,
|
560 |
+
api_name=False,
|
561 |
+
)
|
562 |
+
negative_prompt.submit(
|
563 |
+
fn=randomize_seed_fn,
|
564 |
+
inputs=[seed, randomize_seed],
|
565 |
+
outputs=seed,
|
566 |
+
queue=False,
|
567 |
+
api_name=False,
|
568 |
+
).then(
|
569 |
+
fn=run_full_gradio,
|
570 |
+
inputs=inputs,
|
571 |
+
outputs=result,
|
572 |
+
api_name=False,
|
573 |
+
)
|
574 |
+
run_button.click(
|
575 |
+
fn=randomize_seed_fn,
|
576 |
+
inputs=[seed, randomize_seed],
|
577 |
+
outputs=seed,
|
578 |
+
queue=False,
|
579 |
+
api_name=False,
|
580 |
+
).then(
|
581 |
+
fn=run_full_gradio,
|
582 |
+
inputs=inputs,
|
583 |
+
outputs=result,
|
584 |
+
api_name=False,
|
585 |
+
)
|
586 |
+
|
587 |
+
# API interconnections
|
588 |
+
api_fullexp_run_button.click(
|
589 |
+
fn=run_full_api,
|
590 |
+
inputs=[api_fullexp_image_url, api_fullexp_user_prompt],
|
591 |
+
outputs=[api_fullexp_text_search_url, api_fullexp_visual_search_url, api_fullexp_logs],
|
592 |
+
api_name="full_experience",
|
593 |
+
)
|
594 |
+
api_caponly_run_button.click(
|
595 |
+
fn=run_caponly_api,
|
596 |
+
inputs=[api_caponly_image_url],
|
597 |
+
outputs=[api_caponly_text_search_url, api_caponly_logs],
|
598 |
+
api_name="caption_only",
|
599 |
)
|
600 |
|
601 |
+
if __name__ == "__main__":
|
602 |
+
demo.queue(max_size=20).launch()
|
requirements.txt
CHANGED
@@ -3,4 +3,10 @@ diffusers
|
|
3 |
invisible_watermark
|
4 |
torch
|
5 |
transformers
|
6 |
-
xformers
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
invisible_watermark
|
4 |
torch
|
5 |
transformers
|
6 |
+
xformers
|
7 |
+
flash-attn
|
8 |
+
bitsandbytes
|
9 |
+
azure-core
|
10 |
+
azure-storage-blob
|
11 |
+
azure-identity
|
12 |
+
einops
|
style.css
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#component-0{
|
2 |
+
max-width: 900px;
|
3 |
+
margin: 0 auto;
|
4 |
+
}
|
5 |
+
|
6 |
+
#description, h1 {
|
7 |
+
text-align: center;
|
8 |
+
}
|
9 |
+
|
10 |
+
#duplicate-button {
|
11 |
+
margin: auto;
|
12 |
+
color: #fff;
|
13 |
+
background: #1565c0;
|
14 |
+
border-radius: 100vh;
|
15 |
+
}
|
16 |
+
|