Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -21,14 +21,11 @@ from transformers import (
|
|
| 21 |
GemmaTokenizer
|
| 22 |
)
|
| 23 |
|
|
|
|
| 24 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 25 |
model_repo_id = "Alpha-VLLM/Lumina-Image-2.0"
|
| 26 |
transformer_repo_id = "benjamin-paine/Lumina-Image-2.0" # Temporarily fixed, change when main repo gets updated
|
| 27 |
-
|
| 28 |
-
if torch.cuda.is_available():
|
| 29 |
-
torch_dtype = torch.bfloat16
|
| 30 |
-
else:
|
| 31 |
-
torch_dtype = torch.float32
|
| 32 |
|
| 33 |
###
|
| 34 |
transformer = Lumina2Transformer2DModel.from_pretrained(transformer_repo_id, subfolder="transformer")
|
|
@@ -60,6 +57,9 @@ def infer(
|
|
| 60 |
height=1024,
|
| 61 |
guidance_scale=4.0,
|
| 62 |
num_inference_steps=30,
|
|
|
|
|
|
|
|
|
|
| 63 |
progress=gr.Progress(track_tqdm=True),
|
| 64 |
):
|
| 65 |
if randomize_seed:
|
|
@@ -75,6 +75,9 @@ def infer(
|
|
| 75 |
width=width,
|
| 76 |
height=height,
|
| 77 |
generator=generator,
|
|
|
|
|
|
|
|
|
|
| 78 |
).images[0]
|
| 79 |
|
| 80 |
return image, seed
|
|
@@ -98,7 +101,7 @@ with gr.Blocks(css=css) as demo:
|
|
| 98 |
prompt = gr.Text(
|
| 99 |
label="Prompt",
|
| 100 |
show_label=False,
|
| 101 |
-
max_lines=
|
| 102 |
placeholder="Enter your prompt",
|
| 103 |
container=False,
|
| 104 |
)
|
|
@@ -108,21 +111,29 @@ with gr.Blocks(css=css) as demo:
|
|
| 108 |
result = gr.Image(label="Result", show_label=False)
|
| 109 |
|
| 110 |
with gr.Accordion("Advanced Settings", open=False):
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
|
|
|
| 126 |
|
| 127 |
with gr.Row():
|
| 128 |
width = gr.Slider(
|
|
@@ -158,6 +169,22 @@ with gr.Blocks(css=css) as demo:
|
|
| 158 |
value=30,
|
| 159 |
)
|
| 160 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 161 |
gr.Examples(examples=examples, inputs=[prompt], outputs=[result, seed], fn=infer, cache_examples=True, cache_mode="lazy")
|
| 162 |
|
| 163 |
gr.on(
|
|
@@ -172,6 +199,9 @@ with gr.Blocks(css=css) as demo:
|
|
| 172 |
height,
|
| 173 |
guidance_scale,
|
| 174 |
num_inference_steps,
|
|
|
|
|
|
|
|
|
|
| 175 |
],
|
| 176 |
outputs=[result, seed],
|
| 177 |
)
|
|
|
|
| 21 |
GemmaTokenizer
|
| 22 |
)
|
| 23 |
|
| 24 |
+
default_system_prompt = "You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts."
|
| 25 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 26 |
model_repo_id = "Alpha-VLLM/Lumina-Image-2.0"
|
| 27 |
transformer_repo_id = "benjamin-paine/Lumina-Image-2.0" # Temporarily fixed, change when main repo gets updated
|
| 28 |
+
torch_dtype = torch.float32
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
|
| 30 |
###
|
| 31 |
transformer = Lumina2Transformer2DModel.from_pretrained(transformer_repo_id, subfolder="transformer")
|
|
|
|
| 57 |
height=1024,
|
| 58 |
guidance_scale=4.0,
|
| 59 |
num_inference_steps=30,
|
| 60 |
+
system_prompt=default_system_prompt,
|
| 61 |
+
cfg_normalization=True,
|
| 62 |
+
cfg_trunc_ratio=1.0,
|
| 63 |
progress=gr.Progress(track_tqdm=True),
|
| 64 |
):
|
| 65 |
if randomize_seed:
|
|
|
|
| 75 |
width=width,
|
| 76 |
height=height,
|
| 77 |
generator=generator,
|
| 78 |
+
system_prompt=system_prompt,
|
| 79 |
+
cfg_normalization=cfg_normalization,
|
| 80 |
+
cfg_trunc_ratio=cfg_trunc_ratio
|
| 81 |
).images[0]
|
| 82 |
|
| 83 |
return image, seed
|
|
|
|
| 101 |
prompt = gr.Text(
|
| 102 |
label="Prompt",
|
| 103 |
show_label=False,
|
| 104 |
+
max_lines=4,
|
| 105 |
placeholder="Enter your prompt",
|
| 106 |
container=False,
|
| 107 |
)
|
|
|
|
| 111 |
result = gr.Image(label="Result", show_label=False)
|
| 112 |
|
| 113 |
with gr.Accordion("Advanced Settings", open=False):
|
| 114 |
+
with gr.Row():
|
| 115 |
+
system_prompt = gr.Text(
|
| 116 |
+
label="System Prompt",
|
| 117 |
+
max_lines=4,
|
| 118 |
+
value=default_system_prompt
|
| 119 |
+
)
|
| 120 |
+
|
| 121 |
+
negative_prompt = gr.Text(
|
| 122 |
+
label="Negative prompt",
|
| 123 |
+
max_lines=4,
|
| 124 |
+
placeholder="Enter a negative prompt",
|
| 125 |
+
)
|
| 126 |
|
| 127 |
+
with gr.Row():
|
| 128 |
+
seed = gr.Slider(
|
| 129 |
+
label="Seed",
|
| 130 |
+
minimum=0,
|
| 131 |
+
maximum=MAX_SEED,
|
| 132 |
+
step=1,
|
| 133 |
+
value=0,
|
| 134 |
+
)
|
| 135 |
+
|
| 136 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
| 137 |
|
| 138 |
with gr.Row():
|
| 139 |
width = gr.Slider(
|
|
|
|
| 169 |
value=30,
|
| 170 |
)
|
| 171 |
|
| 172 |
+
with gr.Row():
|
| 173 |
+
cfg_normalization = gr.Checkbox(
|
| 174 |
+
label="CFG Normalization",
|
| 175 |
+
value=True
|
| 176 |
+
)
|
| 177 |
+
cfg_trunc_ratio = gr.Slider(
|
| 178 |
+
label="CFG Truncation Ratio",
|
| 179 |
+
minimum=0.0,
|
| 180 |
+
maximum=1.0,
|
| 181 |
+
step=0.01,
|
| 182 |
+
value=1.0
|
| 183 |
+
)
|
| 184 |
+
|
| 185 |
+
with gr.Row():
|
| 186 |
+
|
| 187 |
+
|
| 188 |
gr.Examples(examples=examples, inputs=[prompt], outputs=[result, seed], fn=infer, cache_examples=True, cache_mode="lazy")
|
| 189 |
|
| 190 |
gr.on(
|
|
|
|
| 199 |
height,
|
| 200 |
guidance_scale,
|
| 201 |
num_inference_steps,
|
| 202 |
+
system_prompt,
|
| 203 |
+
cfg_normalization,
|
| 204 |
+
cfg_trunc_ratio,
|
| 205 |
],
|
| 206 |
outputs=[result, seed],
|
| 207 |
)
|