Spaces:
Running
on
Zero
Running
on
Zero
File size: 8,778 Bytes
7f4c99b 06529b5 7f4c99b 06529b5 7f4c99b aa0cb15 7f4c99b fc39399 7f4c99b 0fb3ed4 02f82c1 0fb3ed4 75ba08b 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b c452fe5 7f4c99b c452fe5 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 80ddbec ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b ce80858 7f4c99b 2b7d18b 7f4c99b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 |
import os
import subprocess
import sys
import io
import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import Flux2Pipeline, Flux2Transformer2DModel
from diffusers import BitsAndBytesConfig as DiffBitsAndBytesConfig
from optimization import optimize_pipeline_
import requests
from PIL import Image
import json
dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024
def remote_text_encoder(prompts):
from gradio_client import Client
client = Client("multimodalart/mistral-text-encoder")
result = client.predict(
prompt=prompts,
api_name="/encode_text"
)
prompt_embeds = torch.load(result[0])
return prompt_embeds
# Load model
repo_id = "black-forest-labs/FLUX.2-dev"
dit = Flux2Transformer2DModel.from_pretrained(
repo_id,
subfolder="transformer",
torch_dtype=torch.bfloat16
)
pipe = Flux2Pipeline.from_pretrained(
repo_id,
text_encoder=None,
transformer=dit,
torch_dtype=torch.bfloat16
)
pipe.to("cuda")
pipe.transformer.set_attention_backend("_flash_3_hub")
optimize_pipeline_(
pipe,
image=[Image.new("RGB", (1024, 1024))],
prompt_embeds = remote_text_encoder("prompt").to("cuda"),
guidance_scale=2.5,
width=1024,
height=1024,
num_inference_steps=1
)
def get_duration(prompt, input_images=None, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=50, guidance_scale=2.5, progress=gr.Progress(track_tqdm=True)):
num_images = 0 if input_images is None else len(input_images)
step_duration = 1 + 0.7 * num_images
return max(65, num_inference_steps * step_duration + 10)
@spaces.GPU(duration=get_duration)
def infer(prompt, input_images=None, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=50, guidance_scale=2.5, progress=gr.Progress(track_tqdm=True)):
if randomize_seed:
seed = random.randint(0, MAX_SEED)
# Get prompt embeddings from remote text encoder
progress(0.1, desc="ํ๋กฌํํธ ์ธ์ฝ๋ฉ ์ค...")
prompt_embeds = remote_text_encoder(prompt).to("cuda")
# Prepare image list (convert None or empty gallery to None)
image_list = None
if input_images is not None and len(input_images) > 0:
image_list = []
for item in input_images:
image_list.append(item[0])
# Generate image
progress(0.3, desc="์ด๋ฏธ์ง ์์ฑ ์ค...")
generator = torch.Generator(device=device).manual_seed(seed)
image = pipe(
prompt_embeds=prompt_embeds,
image=image_list,
width=width,
height=height,
num_inference_steps=num_inference_steps,
guidance_scale=guidance_scale,
generator=generator,
).images[0]
return image, seed
examples = [
["๊ฑฐ์ค ํ์ ์์ ๊ฝ๋ณ์ ๋ง๋ค์ด ์ฃผ์ธ์. ๊ฝ๋ณ์ ์์์ #02eb3c ์์์์ ์์ํ์ฌ #edfa3c๋ก ๋๋๋ ๊ทธ๋ผ๋ฐ์ด์
์
๋๋ค. ๊ฝ๋ณ ์์ ๊ฝ๋ค์ #ff0088 ์์์
๋๋ค"],
["๋ฒ ๋ฅผ๋ฆฐ TV ํ์(Fernsehturm)์ ์ ์ฒด ๊ตฌ์กฐ๋ฅผ ์ง๋ฉด ๊ธฐ์ด๋ถํฐ ์ํ
๋ ๋๊น์ง ๋ณด์ฌ์ฃผ๋ ์ฌ์ง์ฒ๋ผ ์ฌ์ค์ ์ธ ์ธํฌ๊ทธ๋ํฝ, ์ฝํฌ๋ฆฌํธ ์ถ, ๊ธ์ ๊ตฌ์ฒด, ์ํ
๋ ์ฒจํ์ ํฌํจํ ์ ์ฒด ๊ตฌ์กฐ๊ฐ ๋ณด์ด๋ ์์ง ์ ์ฒด ๋ทฐ. ์์ง์ ์ธ ๊ตฌ์ฒด๋ฅผ ์ฌ๋ ค๋ค๋ณด๋ ์ฝ๊ฐ์ ์์ชฝ ์๊ทผ๊ฐ ๊ฐ๋, ๊นจ๋ํ ํฐ์ ๋ฐฐ๊ฒฝ์ ์๋ฒฝํ๊ฒ ์ค์ ๋ฐฐ์น. ์์ ์ํ ์ฐ๊ฒฐ์ ์ด ์๋ ์ผ์ชฝ ๋ ์ด๋ธ: ๋งค์ฐ ํฐ ๊ตต์ ์งํ ํ์ ์ซ์(#2D3748)๋ก ๋ '368m' ํ
์คํธ๊ฐ ์ํ
๋ ๋์ ์ ํํ ์์นํ๊ณ ๊ทธ ์๋์ ์์ ๋๋ฌธ์๋ก 'TOTAL HEIGHT'๊ฐ ์์. ๋งค์ฐ ํฐ ๊ตต์ ๊ธ์จ๋ก ๋ '207m' ํ
์คํธ์ ๊ทธ ์๋ ์์ ๋๋ฌธ์๋ก 'TELECAFร'๊ฐ ์์ผ๋ฉฐ, ์ฐ๊ฒฐ์ ์ด ์ฐฝ๋ฌธ ๋์ด์ ๊ตฌ์ฒด์ ์ ํํ ๋ฟ์ ์์. ๊ตฌ์ฒด์ ์ ๋์ ๋ฟ๋ ์ํ ์ฐ๊ฒฐ์ ์ด ์๋ ์ค๋ฅธ์ชฝ ๋ ์ด๋ธ: ๋งค์ฐ ํฐ ๊ตต์ ์งํ ํ์ ์ซ์๋ก ๋ '32m' ํ
์คํธ์ ๊ทธ ์๋ ์์ ๋๋ฌธ์๋ก 'SPHERE DIAMETER'๊ฐ ์์. ์ธ ๊ฐ์ ๊ท ํ ์กํ ์ด๋ก ๋ฐฐ์ด๋ ํ๋จ ์น์
: ์ผ์ชฝ - ๋งค์ฐ ๊ตต์ ์งํ ํ์์ ํฐ ํ
์คํธ '986'๊ณผ ๊ทธ ์๋ ๋๋ฌธ์๋ก 'STEPS'. ์ค์ - ๊ตต์ ๋๋ฌธ์๋ก 'BERLIN TV TOWER'์ ๊ทธ ์๋ ๊ฐ๋ฒผ์ด ๋ฌด๊ฒ๋ก 'FERNSEHTURM'. ์ค๋ฅธ์ชฝ - ๊ตต์ ๋๋ฌธ์๋ก 'INAUGURATED'์ ๊ทธ ์๋ 'OCTOBER 3, 1969'. ๋ชจ๋ ํ์ดํฌ๊ทธ๋ํผ๋ ํ๋์ ์ธ ์ฐ์ธ๋ฆฌํ ํฐํธ(Inter ๋๋ Helvetica ๊ฐ์), ์์ #2D3748, ๊นจ๋ํ๊ณ ๋ฏธ๋๋ฉํ ๊ธฐ์ ๋ค์ด์ด๊ทธ๋จ ์คํ์ผ. ์ํ ์ฐ๊ฒฐ์ ์ ์๊ณ ์ ํํ๋ฉฐ ๋ช
ํํ๊ฒ ๋ณด์ด๊ณ ํ์ ๊ตฌ์กฐ์ ์ ํํ ํด๋น ์ธก์ ์ง์ ์ ๋ฟ์ ์์. ๋์ด์ ์
์ฅํจ์ ๋๋ ์ ์๋ ์ญ๋์ ์ธ ๋ฎ์ ๊ฐ๋ ์๊ทผ๊ฐ์ด ์๋ ์ ๋ฌธ์ ์ธ ๊ฑด์ถ ์
๋ฉด๋ ๋ฏธํ, ์๋ฒฝํ ์๊ฐ์ ๊ณ์ธต ๊ตฌ์กฐ๋ฅผ ๊ฐ์ง ํฌ์คํฐ๊ธ ์ธํฌ๊ทธ๋ํฝ ๋์์ธ."],
["๋น์ค๋ ์ ๊ธ์์ ๋ฐ๋๋ ์ ์๋ ํผ์ ํ๊ณ ์๋ ํ ๋ป ์ ์ ์นดํผ๋ฐ๋ผ, ํด๋ก์ฆ์
์ฌ์ง"],
["ํตํตํ ์ฃผํฉ์ ๊ณ ์์ด์ ์นด์์ด ๋ค์ด์ปท ์คํฐ์ปค, ํฌ๊ณ ๋ฐ์ง์ด๋ ๋๊ณผ ์ธ์ฌํ๋ฉฐ ๋ฐ์ ์ฌ๋ฆฐ ํ๋ณตํ ๋ฏธ์์ ํํธ ๋ชจ์์ ๋ถํ ์ฝ๊ฐ ์์ต๋๋ค. ๋์์ธ์ ๊ฒ์์ ์ค๊ณฝ์ ๊ณผ ๋ถํ ๋ณผ์ด ์๋ ๋ถ๋๋ฌ์ด ๊ทธ๋ผ๋ฐ์ด์
์์์ด ์๋ ๋ถ๋๋ฌ์ด ๋ฅ๊ทผ ์ ์ด ์์ด์ผ ํฉ๋๋ค."],
]
examples_images = [
# ["Replace the top of the person from image 1 with the one from image 2", ["person1.webp", "woman2.webp"]],
["์ด๋ฏธ์ง 1์ ์ฌ๋์ด ์ด๋ฏธ์ง 2์ ๊ณ ์์ด๋ฅผ ์ฐ๋ค๋ฌ๊ณ ์๊ณ , ์ด๋ฏธ์ง 3์ ์๊ฐ ๊ทธ๋ค ์์ ์์ต๋๋ค", ["woman1.webp", "cat_window.webp", "bird.webp"]]
]
css="""
#col-container {
margin: 0 auto;
max-width: 620px;
}
"""
with gr.Blocks() as demo:
with gr.Column(elem_id="col-container"):
gr.Markdown(f"""# FLUX.2 [dev]
FLUX.2 [dev]๋ ํ
์คํธ ์ง์์ฌํญ์ ๊ธฐ๋ฐ์ผ๋ก ์ด๋ฏธ์ง๋ฅผ ์์ฑ, ํธ์ง ๋ฐ ๊ฒฐํฉํ ์ ์๋ 32B ํ๋ผ๋ฏธํฐ rectified flow ๋ชจ๋ธ์
๋๋ค [[๋ชจ๋ธ](https://huggingface.co/black-forest-labs/FLUX.2-dev)], [[๋ธ๋ก๊ทธ](https://bfl.ai/blog/flux-2)]
""")
with gr.Accordion("์
๋ ฅ ์ด๋ฏธ์ง (์ ํ์ฌํญ)", open=False):
input_images = gr.Gallery(
label="์
๋ ฅ ์ด๋ฏธ์ง",
type="pil",
columns=3,
rows=1,
)
prompt = gr.Text(
label="ํ๋กฌํํธ",
show_label=False,
lines=10,
max_lines=15,
placeholder="ํ๋กฌํํธ๋ฅผ ์
๋ ฅํ์ธ์",
container=False,
)
run_button = gr.Button("์คํ")
result = gr.Image(label="๊ฒฐ๊ณผ", show_label=False)
with gr.Accordion("๊ณ ๊ธ ์ค์ ", open=False):
seed = gr.Slider(
label="์๋",
minimum=0,
maximum=MAX_SEED,
step=1,
value=0,
)
randomize_seed = gr.Checkbox(label="๋๋ค ์๋", value=True)
with gr.Row():
width = gr.Slider(
label="๋๋น",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
height = gr.Slider(
label="๋์ด",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
with gr.Row():
num_inference_steps = gr.Slider(
label="์ถ๋ก ๋จ๊ณ ์",
minimum=1,
maximum=100,
step=1,
value=30,
)
guidance_scale = gr.Slider(
label="๊ฐ์ด๋์ค ์ค์ผ์ผ",
minimum=0.0,
maximum=10.0,
step=0.1,
value=4,
)
gr.Examples(
examples=examples,
fn=infer,
inputs=[prompt],
outputs=[result, seed],
cache_examples=True,
cache_mode="lazy"
)
gr.Examples(
examples=examples_images,
fn=infer,
inputs=[prompt, input_images],
outputs=[result, seed],
cache_examples=True,
cache_mode="lazy"
)
gr.on(
triggers=[run_button.click, prompt.submit],
fn=infer,
inputs=[prompt, input_images, seed, randomize_seed, width, height, num_inference_steps, guidance_scale],
outputs=[result, seed]
)
demo.launch(css=css) |