| import gradio as gr |
| import numpy as np |
| import torch |
| import spaces |
| from PIL import Image |
|
|
| MAX_SEED = np.iinfo(np.int32).max |
| MAX_IMAGE_SIZE = 2048 |
|
|
| import os |
| hf_token = os.environ.get("HF_TOKEN") |
|
|
| PIPE = None |
| DEPTH_PROC = None |
| DEPTH_MODEL = None |
|
|
|
|
| def load_models(): |
| global PIPE, DEPTH_PROC, DEPTH_MODEL |
| if PIPE is not None: |
| return |
| from diffusers import FluxControlPipeline |
| from transformers import AutoImageProcessor, DepthAnythingForDepthEstimation |
| print("Loading FLUX.1-Depth-dev ...", flush=True) |
| PIPE = FluxControlPipeline.from_pretrained( |
| "black-forest-labs/FLUX.1-Depth-dev", |
| torch_dtype=torch.bfloat16, |
| token=hf_token, |
| low_cpu_mem_usage=True, |
| ).to("cuda") |
| PIPE.enable_sequential_cpu_offload() |
| PIPE.enable_vae_tiling() |
| PIPE.enable_vae_slicing() |
| print("FLUX loaded.", flush=True) |
| depth_id = "LiheYoung/depth-anything-large-hf" |
| print("Loading DepthAnything ...", flush=True) |
| DEPTH_PROC = AutoImageProcessor.from_pretrained(depth_id, token=hf_token) |
| DEPTH_MODEL = DepthAnythingForDepthEstimation.from_pretrained(depth_id, token=hf_token).to("cuda") |
| print("DepthAnything loaded.", flush=True) |
|
|
|
|
| def to_depth(image_pil): |
| inputs = DEPTH_PROC(images=image_pil, return_tensors="pt").to("cuda") |
| with torch.no_grad(): |
| out = DEPTH_MODEL(**inputs).predicted_depth |
| depth = out.squeeze().cpu().float().numpy() |
| depth = (depth - depth.min()) / (depth.max() - depth.min() + 1e-8) |
| depth = (depth * 255.0).astype("uint8") |
| return Image.fromarray(depth).convert("RGB") |
|
|
|
|
| @spaces.GPU |
| def infer(control_image, prompt, seed=42, randomize_seed=False, width=1024, height=1024, |
| guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)): |
| try: |
| load_models() |
| if randomize_seed: |
| seed = int(np.random.randint(0, MAX_SEED)) |
| control_image = to_depth(control_image.convert("RGB")) |
| image = PIPE( |
| prompt=prompt, |
| control_image=control_image, |
| height=height, |
| width=width, |
| num_inference_steps=num_inference_steps, |
| guidance_scale=guidance_scale, |
| generator=torch.Generator().manual_seed(seed), |
| ).images[0] |
| return image, seed |
| except Exception as e: |
| import traceback |
| traceback.print_exc() |
| raise |
|
|
|
|
| examples = [ |
| "a tiny astronaut hatching from an egg on the moon", |
| "a cat holding a sign that says hello world", |
| "an anime illustration of a wiener schnitzel", |
| ] |
|
|
| css = """#col-container { margin: 0 auto; max-width: 520px; }""" |
|
|
| with gr.Blocks(css=css) as demo: |
| with gr.Column(elem_id="col-container"): |
| gr.Markdown( |
| "# FLUX.1 Depth [dev]\n" |
| "12B param rectified flow transformer structural conditioning, guidance-distilled. " |
| "Non-commercial license." |
| ) |
| control_image = gr.Image(label="Upload the image for control", type="pil") |
| with gr.Row(): |
| prompt = gr.Text(label="Prompt", show_label=False, max_lines=1, |
| placeholder="Enter your prompt", container=False) |
| run_button = gr.Button("Run", scale=0) |
| result = gr.Image(label="Result", show_label=False) |
| with gr.Accordion("Advanced Settings", open=False): |
| seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0) |
| randomize_seed = gr.Checkbox(label="Randomize seed", value=True) |
| with gr.Row(): |
| width = gr.Slider(label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024) |
| height = gr.Slider(label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024) |
| with gr.Row(): |
| guidance_scale = gr.Slider(label="Guidance Scale", minimum=1, maximum=30, step=0.5, value=10) |
| num_inference_steps = gr.Slider(label="Number of inference steps", minimum=1, maximum=50, step=1, value=28) |
|
|
| gr.on( |
| triggers=[run_button.click, prompt.submit], |
| fn=infer, |
| inputs=[control_image, prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps], |
| outputs=[result, seed], |
| ) |
|
|
| demo.launch() |
|
|
