Spaces:
Running
on
Zero
Running
on
Zero
File size: 6,897 Bytes
0b7c365 |
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 |
import gradio as gr
import torch
from diffusers import HiDreamImagePipeline
from transformers import PreTrainedTokenizerFast, LlamaForCausalLM
import random
import spaces
import numpy as np
# Set data type
dtype = torch.bfloat16
device = "cpu" # Initial device for model loading; inference will use GPU
# Load tokenizer and text encoder for Llama
try:
tokenizer_4 = PreTrainedTokenizerFast.from_pretrained("meta-llama/Meta-Llama-3.1-8B-Instruct")
text_encoder_4 = LlamaForCausalLM.from_pretrained(
"meta-llama/Meta-Llama-3.1-8B-Instruct",
output_hidden_states=True,
output_attentions=True,
torch_dtype=dtype,
).to(device)
except Exception as e:
raise Exception(f"Failed to load Llama model: {e}. Ensure you have access to 'meta-llama/Meta-Llama-3.1-8B-Instruct' and are logged in via `huggingface-cli login`.")
# Load the HiDreamImagePipeline
try:
pipe = HiDreamImagePipeline.from_pretrained(
"HiDream-ai/HiDream-I1-Full",
tokenizer_4=tokenizer_4,
text_encoder_4=text_encoder_4,
torch_dtype=dtype,
).to(device)
pipe.enable_model_cpu_offload() # Offload to CPU when not in use, critical for Spaces
except Exception as e:
raise Exception(f"Failed to load HiDreamImagePipeline: {e}. Ensure you have access to 'HiDream-ai/HiDream-I1-Full'.")
# Define maximum values
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048
# Inference function with GPU access
@spaces.GPU()
def infer(prompt, negative_prompt="", seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=50, guidance_scale=5.0, progress=gr.Progress(track_tqdm=True)):
# Ensure the model is on GPU for inference
pipe.to("cuda")
try:
if randomize_seed:
seed = random.randint(0, MAX_SEED)
generator = torch.Generator("cuda").manual_seed(seed)
# Generate the image
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
height=height,
width=width,
num_inference_steps=num_inference_steps,
guidance_scale=guidance_scale,
generator=generator,
output_type="pil",
).images[0]
# Clear GPU memory
torch.cuda.empty_cache()
return image, seed
finally:
# Move model back to CPU to free GPU memory
pipe.to("cpu")
torch.cuda.empty_cache()
# Define examples
examples = [
["A cat holding a sign that says \"Hi-Dreams.ai\".", ""],
["A futuristic cityscape with flying cars.", "blurry, low quality"],
["A serene landscape with mountains and a lake.", ""],
]
# CSS styling
css = """
#col-container {
margin: 0 auto;
max-width: 960px;
}
.generate-btn {
background: linear-gradient(90deg, #4B79A1 0%, #283E51 100%) !important;
border: none !important;
color: white !important;
}
.generate-btn:hover {
transform: translateY(-2px);
box-shadow: 0 5px 15px rgba(0,0,0,0.2);
}
"""
# Create Gradio interface
with gr.Blocks(css=css) as app:
gr.HTML("<center><h1>HiDreamImage Generator</h1></center>")
with gr.Column(elem_id="col-container"):
with gr.Row():
with gr.Column():
with gr.Row():
text_prompt = gr.Textbox(
label="Prompt",
placeholder="Enter a prompt here",
lines=3,
elem_id="prompt-text-input"
)
with gr.Row():
with gr.Accordion("Advanced Settings", open=False):
with gr.Row():
width = gr.Slider(
label="Width",
value=1024,
minimum=64,
maximum=MAX_IMAGE_SIZE,
step=8
)
height = gr.Slider(
label="Height",
value=1024,
minimum=64,
maximum=MAX_IMAGE_SIZE,
step=8
)
with gr.Row():
steps = gr.Slider(
label="Inference Steps",
value=50,
minimum=1,
maximum=100,
step=1
)
cfg = gr.Slider(
label="Guidance Scale",
value=5.0,
minimum=1,
maximum=20,
step=0.5
)
with gr.Row():
seed = gr.Slider(
label="Seed",
value=42,
minimum=0,
maximum=MAX_SEED,
step=1
)
randomize_seed = gr.Checkbox(
label="Randomize Seed",
value=True
)
with gr.Row():
negative_prompt = gr.Textbox(
label="Negative Prompt",
placeholder="Enter what to avoid (optional)",
lines=2
)
with gr.Row():
text_button = gr.Button(
"✨ Generate Image",
variant='primary',
elem_classes=["generate-btn"]
)
with gr.Column():
with gr.Row():
image_output = gr.Image(
type="pil",
label="Generated Image",
elem_id="gallery"
)
seed_output = gr.Textbox(
label="Seed Used",
interactive=False
)
with gr.Column():
gr.Examples(
examples=examples,
inputs=[text_prompt, negative_prompt],
)
# Connect the button and textbox submit to the inference function
gr.on(
triggers=[text_button.click, text_prompt.submit],
fn=infer,
inputs=[text_prompt, negative_prompt, seed, randomize_seed, width, height, steps, cfg],
outputs=[image_output, seed_output]
)
# Launch the app
app.launch(share=True) |