Spaces:
Running
on
Zero
Running
on
Zero
feat: image generation feature
Browse files
app.py
CHANGED
@@ -4,9 +4,13 @@ from PIL import Image
|
|
4 |
import gradio as gr
|
5 |
|
6 |
from v2 import V2UI
|
7 |
-
from diffusion import ImageGenerator
|
8 |
from output import UpsamplingOutput
|
9 |
-
from utils import
|
|
|
|
|
|
|
|
|
10 |
|
11 |
|
12 |
NORMALIZE_RATING_TAG = {
|
@@ -53,11 +57,7 @@ def elapsed_time_format(elapsed_time: float) -> str:
|
|
53 |
def parse_upsampling_output(
|
54 |
upsampler: Callable[..., UpsamplingOutput],
|
55 |
):
|
56 |
-
def _parse_upsampling_output(*args) -> tuple[
|
57 |
-
str,
|
58 |
-
str,
|
59 |
-
dict,
|
60 |
-
]:
|
61 |
output = upsampler(*args)
|
62 |
|
63 |
print(output)
|
@@ -68,54 +68,14 @@ def parse_upsampling_output(
|
|
68 |
gr.update(
|
69 |
interactive=True,
|
70 |
),
|
|
|
|
|
|
|
71 |
)
|
72 |
|
73 |
return _parse_upsampling_output
|
74 |
|
75 |
|
76 |
-
def image_generation_config_ui():
|
77 |
-
with gr.Accordion(label="Image generation config", open=False) as accordion:
|
78 |
-
image_size = gr.Radio(
|
79 |
-
label="Image size",
|
80 |
-
choices=list(IMAGE_SIZE_OPTIONS.keys()),
|
81 |
-
value=list(IMAGE_SIZE_OPTIONS.keys())[3], # tall
|
82 |
-
)
|
83 |
-
|
84 |
-
quality_tags = gr.Textbox(
|
85 |
-
label="Quality tags",
|
86 |
-
placeholder=QUALITY_TAGS["default"],
|
87 |
-
value=QUALITY_TAGS["default"],
|
88 |
-
)
|
89 |
-
negative_prompt = gr.Textbox(
|
90 |
-
label="Negative prompt",
|
91 |
-
placeholder=NEGATIVE_PROMPT["default"],
|
92 |
-
value=NEGATIVE_PROMPT["default"],
|
93 |
-
)
|
94 |
-
|
95 |
-
num_inference_steps = gr.Slider(
|
96 |
-
label="Num inference steps",
|
97 |
-
minimum=20,
|
98 |
-
maximum=30,
|
99 |
-
step=1,
|
100 |
-
value=25,
|
101 |
-
)
|
102 |
-
guidance_scale = gr.Slider(
|
103 |
-
label="Guidance scale",
|
104 |
-
minimum=0.0,
|
105 |
-
maximum=10.0,
|
106 |
-
step=0.5,
|
107 |
-
value=7.0,
|
108 |
-
)
|
109 |
-
|
110 |
-
return accordion, [
|
111 |
-
image_size,
|
112 |
-
quality_tags,
|
113 |
-
negative_prompt,
|
114 |
-
num_inference_steps,
|
115 |
-
guidance_scale,
|
116 |
-
]
|
117 |
-
|
118 |
-
|
119 |
def description_ui():
|
120 |
gr.Markdown(
|
121 |
"""
|
@@ -129,7 +89,7 @@ def main():
|
|
129 |
v2 = V2UI()
|
130 |
|
131 |
print("Loading diffusion model...")
|
132 |
-
|
133 |
print("Loaded.")
|
134 |
|
135 |
with gr.Blocks() as ui:
|
@@ -140,12 +100,18 @@ def main():
|
|
140 |
v2.ui()
|
141 |
|
142 |
with gr.Column():
|
143 |
-
|
|
|
|
|
|
|
|
|
|
|
144 |
|
145 |
elapsed_time_md = gr.Markdown(label="Elapsed time", value="")
|
146 |
|
147 |
generate_image_btn = gr.Button(
|
148 |
value="Generate image with this prompt!",
|
|
|
149 |
)
|
150 |
|
151 |
accordion, image_generation_config_components = (
|
@@ -153,11 +119,11 @@ def main():
|
|
153 |
)
|
154 |
|
155 |
output_image = gr.Gallery(
|
156 |
-
label="
|
|
|
157 |
columns=1,
|
158 |
preview=True,
|
159 |
-
|
160 |
-
visible=False,
|
161 |
)
|
162 |
|
163 |
gr.Examples(
|
@@ -216,6 +182,15 @@ def main():
|
|
216 |
"long",
|
217 |
"lax",
|
218 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
219 |
[
|
220 |
"honkai: star rail",
|
221 |
"silver wolf (honkai: star rail)",
|
@@ -245,7 +220,13 @@ def main():
|
|
245 |
inputs=[
|
246 |
*v2.get_inputs(),
|
247 |
],
|
248 |
-
outputs=[output_text, elapsed_time_md, generate_image_btn],
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
)
|
250 |
|
251 |
ui.launch()
|
|
|
4 |
import gradio as gr
|
5 |
|
6 |
from v2 import V2UI
|
7 |
+
from diffusion import ImageGenerator, image_generation_config_ui
|
8 |
from output import UpsamplingOutput
|
9 |
+
from utils import (
|
10 |
+
PEOPLE_TAGS,
|
11 |
+
gradio_copy_text,
|
12 |
+
COPY_ACTION_JS,
|
13 |
+
)
|
14 |
|
15 |
|
16 |
NORMALIZE_RATING_TAG = {
|
|
|
57 |
def parse_upsampling_output(
|
58 |
upsampler: Callable[..., UpsamplingOutput],
|
59 |
):
|
60 |
+
def _parse_upsampling_output(*args) -> tuple[str, str, dict, dict]:
|
|
|
|
|
|
|
|
|
61 |
output = upsampler(*args)
|
62 |
|
63 |
print(output)
|
|
|
68 |
gr.update(
|
69 |
interactive=True,
|
70 |
),
|
71 |
+
gr.update(
|
72 |
+
interactive=True,
|
73 |
+
),
|
74 |
)
|
75 |
|
76 |
return _parse_upsampling_output
|
77 |
|
78 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
def description_ui():
|
80 |
gr.Markdown(
|
81 |
"""
|
|
|
89 |
v2 = V2UI()
|
90 |
|
91 |
print("Loading diffusion model...")
|
92 |
+
image_generator = ImageGenerator()
|
93 |
print("Loaded.")
|
94 |
|
95 |
with gr.Blocks() as ui:
|
|
|
100 |
v2.ui()
|
101 |
|
102 |
with gr.Column():
|
103 |
+
with gr.Group():
|
104 |
+
output_text = gr.TextArea(label="Output tags", interactive=False)
|
105 |
+
copy_btn = gr.Button(
|
106 |
+
value="Copy to clipboard",
|
107 |
+
interactive=False,
|
108 |
+
)
|
109 |
|
110 |
elapsed_time_md = gr.Markdown(label="Elapsed time", value="")
|
111 |
|
112 |
generate_image_btn = gr.Button(
|
113 |
value="Generate image with this prompt!",
|
114 |
+
interactive=False,
|
115 |
)
|
116 |
|
117 |
accordion, image_generation_config_components = (
|
|
|
119 |
)
|
120 |
|
121 |
output_image = gr.Gallery(
|
122 |
+
label="Generated image",
|
123 |
+
show_label=True,
|
124 |
columns=1,
|
125 |
preview=True,
|
126 |
+
visible=True,
|
|
|
127 |
)
|
128 |
|
129 |
gr.Examples(
|
|
|
182 |
"long",
|
183 |
"lax",
|
184 |
],
|
185 |
+
[
|
186 |
+
"honkai: star rail",
|
187 |
+
"firefly (honkai: star rail)",
|
188 |
+
"1girl, solo",
|
189 |
+
"sfw",
|
190 |
+
"tall",
|
191 |
+
"medium",
|
192 |
+
"lax",
|
193 |
+
],
|
194 |
[
|
195 |
"honkai: star rail",
|
196 |
"silver wolf (honkai: star rail)",
|
|
|
220 |
inputs=[
|
221 |
*v2.get_inputs(),
|
222 |
],
|
223 |
+
outputs=[output_text, elapsed_time_md, copy_btn, generate_image_btn],
|
224 |
+
)
|
225 |
+
copy_btn.click(gradio_copy_text, inputs=[output_text], js=COPY_ACTION_JS)
|
226 |
+
generate_image_btn.click(
|
227 |
+
image_generator.generate,
|
228 |
+
inputs=[output_text, *image_generation_config_components],
|
229 |
+
outputs=[output_image],
|
230 |
)
|
231 |
|
232 |
ui.launch()
|
diffusion.py
CHANGED
@@ -19,12 +19,61 @@ except ImportError:
|
|
19 |
return lambda x: x
|
20 |
|
21 |
|
22 |
-
|
|
|
23 |
|
24 |
|
25 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
26 |
|
27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
class ImageGenerator:
|
29 |
pipe: StableDiffusionXLPipeline
|
30 |
|
@@ -56,12 +105,18 @@ class ImageGenerator:
|
|
56 |
def generate(
|
57 |
self,
|
58 |
prompt: str,
|
|
|
|
|
59 |
negative_prompt: str = NEGATIVE_PROMPT["default"], # Light v3.1
|
60 |
-
height: int = 1152,
|
61 |
-
width: int = 896,
|
62 |
num_inference_steps: int = 25,
|
63 |
guidance_scale: float = 7.0,
|
64 |
) -> Image.Image:
|
|
|
|
|
|
|
|
|
65 |
print("prompt", prompt)
|
66 |
print("negative_prompt", negative_prompt)
|
67 |
print("height", height)
|
|
|
19 |
return lambda x: x
|
20 |
|
21 |
|
22 |
+
import gradio as gr
|
23 |
+
from utils import NEGATIVE_PROMPT, IMAGE_SIZE_OPTIONS, QUALITY_TAGS, IMAGE_SIZES
|
24 |
|
25 |
|
26 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
27 |
|
28 |
|
29 |
+
def image_generation_config_ui():
|
30 |
+
with gr.Accordion(label="Image generation config", open=False) as accordion:
|
31 |
+
image_size = gr.Radio(
|
32 |
+
label="Image size",
|
33 |
+
choices=list(IMAGE_SIZE_OPTIONS.keys()),
|
34 |
+
value=list(IMAGE_SIZE_OPTIONS.keys())[3],
|
35 |
+
interactive=True,
|
36 |
+
)
|
37 |
+
|
38 |
+
quality_tags = gr.Textbox(
|
39 |
+
label="Quality tags",
|
40 |
+
placeholder=QUALITY_TAGS["default"],
|
41 |
+
value=QUALITY_TAGS["default"],
|
42 |
+
interactive=True,
|
43 |
+
)
|
44 |
+
negative_prompt = gr.Textbox(
|
45 |
+
label="Negative prompt",
|
46 |
+
placeholder=NEGATIVE_PROMPT["default"],
|
47 |
+
value=NEGATIVE_PROMPT["default"],
|
48 |
+
interactive=True,
|
49 |
+
)
|
50 |
+
|
51 |
+
num_inference_steps = gr.Slider(
|
52 |
+
label="Num inference steps",
|
53 |
+
minimum=20,
|
54 |
+
maximum=30,
|
55 |
+
step=1,
|
56 |
+
value=25,
|
57 |
+
interactive=True,
|
58 |
+
)
|
59 |
+
guidance_scale = gr.Slider(
|
60 |
+
label="Guidance scale",
|
61 |
+
minimum=0.0,
|
62 |
+
maximum=10.0,
|
63 |
+
step=0.5,
|
64 |
+
value=7.0,
|
65 |
+
interactive=True,
|
66 |
+
)
|
67 |
+
|
68 |
+
return accordion, [
|
69 |
+
image_size,
|
70 |
+
quality_tags,
|
71 |
+
negative_prompt,
|
72 |
+
num_inference_steps,
|
73 |
+
guidance_scale,
|
74 |
+
]
|
75 |
+
|
76 |
+
|
77 |
class ImageGenerator:
|
78 |
pipe: StableDiffusionXLPipeline
|
79 |
|
|
|
105 |
def generate(
|
106 |
self,
|
107 |
prompt: str,
|
108 |
+
image_size: str = "768x1344",
|
109 |
+
quality_tags: str = QUALITY_TAGS["default"], # Light v3.1
|
110 |
negative_prompt: str = NEGATIVE_PROMPT["default"], # Light v3.1
|
111 |
+
# height: int = 1152,
|
112 |
+
# width: int = 896,
|
113 |
num_inference_steps: int = 25,
|
114 |
guidance_scale: float = 7.0,
|
115 |
) -> Image.Image:
|
116 |
+
width, height = IMAGE_SIZES[image_size]
|
117 |
+
|
118 |
+
prompt = ", ".join([prompt, quality_tags])
|
119 |
+
|
120 |
print("prompt", prompt)
|
121 |
print("negative_prompt", negative_prompt)
|
122 |
print("height", height)
|
utils.py
CHANGED
@@ -1,3 +1,4 @@
|
|
|
|
1 |
from dartrs.v2 import AspectRatioTag, LengthTag, RatingTag, IdentityTag
|
2 |
|
3 |
# from https://huggingface.co/spaces/cagliostrolab/animagine-xl-3.1/blob/main/config.py
|
@@ -59,3 +60,17 @@ PEOPLE_TAGS = [
|
|
59 |
*[f"6+{x}s" for x in ["girl", "boy", "other"]],
|
60 |
"no humans",
|
61 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
from dartrs.v2 import AspectRatioTag, LengthTag, RatingTag, IdentityTag
|
3 |
|
4 |
# from https://huggingface.co/spaces/cagliostrolab/animagine-xl-3.1/blob/main/config.py
|
|
|
60 |
*[f"6+{x}s" for x in ["girl", "boy", "other"]],
|
61 |
"no humans",
|
62 |
]
|
63 |
+
|
64 |
+
|
65 |
+
# ref: https://qiita.com/tregu148/items/fccccbbc47d966dd2fc2
|
66 |
+
def gradio_copy_text(_text: None):
|
67 |
+
gr.Info("Copied!")
|
68 |
+
|
69 |
+
|
70 |
+
COPY_ACTION_JS = """\
|
71 |
+
(inputs, _outputs) => {
|
72 |
+
// inputs is the string value of the input_text
|
73 |
+
if (inputs.trim() !== "") {
|
74 |
+
navigator.clipboard.writeText(inputs);
|
75 |
+
}
|
76 |
+
}"""
|
v2.py
CHANGED
@@ -30,11 +30,6 @@ from utils import ASPECT_RATIO_OPTIONS, RATING_OPTIONS, LENGTH_OPTIONS, IDENTITY
|
|
30 |
HF_TOKEN = os.getenv("HF_TOKEN", None)
|
31 |
|
32 |
ALL_MODELS = {
|
33 |
-
"dart-v2-mixtral-160m-sft-6": {
|
34 |
-
"repo": "p1atdev/dart-v2-mixtral-160m-sft-6",
|
35 |
-
"type": "sft",
|
36 |
-
"class": MixtralModel,
|
37 |
-
},
|
38 |
"dart-v2-mixtral-160m-sft-8": {
|
39 |
"repo": "p1atdev/dart-v2-mixtral-160m-sft-8",
|
40 |
"type": "sft",
|
|
|
30 |
HF_TOKEN = os.getenv("HF_TOKEN", None)
|
31 |
|
32 |
ALL_MODELS = {
|
|
|
|
|
|
|
|
|
|
|
33 |
"dart-v2-mixtral-160m-sft-8": {
|
34 |
"repo": "p1atdev/dart-v2-mixtral-160m-sft-8",
|
35 |
"type": "sft",
|