Spaces:
Sleeping
Sleeping
File size: 10,946 Bytes
d325ede | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 | # pyright: reportMissingImports=false
"""
TwoTh Hugging Face Space
Chained AI pipeline:
1) tencent/Hunyuan3D-2mv for multiview shape generation.
2) tencent/Hunyuan3D-2.1 Paint for PBR texturing.
Returns a single GLB model and JSON stats.
"""
from __future__ import annotations
import os
import shutil
import time
import uuid
from pathlib import Path
from typing import Any, Dict, List, Sequence, Tuple
import gradio as gr
import spaces
import torch
import trimesh
from hy3dgen.shapegen import Hunyuan3DDiTFlowMatchingPipeline
SHAPE_MODEL_ID = os.getenv("SHAPE_MODEL_ID", "tencent/Hunyuan3D-2mv")
SHAPE_SUBFOLDER = os.getenv("SHAPE_SUBFOLDER", "hunyuan3d-dit-v2-mv")
PAINT_MODEL_ID = os.getenv("PAINT_MODEL_ID", "tencent/Hunyuan3D-2.1")
TEMP_ROOT = Path(os.getenv("SPACE_TMP_DIR", "/tmp/twoth-space"))
TEMP_TTL_SECONDS = int(os.getenv("TEMP_TTL_SECONDS", "3600"))
TEMP_ROOT.mkdir(parents=True, exist_ok=True)
_shape_pipeline: Any | None = None
_paint_pipeline: Any | None = None
_paint_pipeline_name = "not_initialized"
_temp_jobs: Dict[str, float] = {}
def _cleanup_temp_jobs() -> None:
now = time.time()
stale_dirs: List[str] = []
for job_dir, created_at in _temp_jobs.items():
if now - created_at > TEMP_TTL_SECONDS:
stale_dirs.append(job_dir)
for stale_dir in stale_dirs:
shutil.rmtree(stale_dir, ignore_errors=True)
_temp_jobs.pop(stale_dir, None)
def _register_temp_job(job_dir: Path) -> None:
_temp_jobs[str(job_dir)] = time.time()
def _load_shape_pipeline() -> Any:
global _shape_pipeline
if _shape_pipeline is not None:
return _shape_pipeline
try:
_shape_pipeline = Hunyuan3DDiTFlowMatchingPipeline.from_pretrained(
SHAPE_MODEL_ID,
subfolder=SHAPE_SUBFOLDER,
use_safetensors=True,
device="cuda",
)
except TypeError:
_shape_pipeline = Hunyuan3DDiTFlowMatchingPipeline.from_pretrained(
SHAPE_MODEL_ID,
subfolder=SHAPE_SUBFOLDER,
)
except Exception:
_shape_pipeline = Hunyuan3DDiTFlowMatchingPipeline.from_pretrained(SHAPE_MODEL_ID)
return _shape_pipeline
def _load_paint_pipeline() -> Tuple[Any, str]:
global _paint_pipeline
global _paint_pipeline_name
if _paint_pipeline is not None:
return _paint_pipeline, _paint_pipeline_name
try:
from hy3dpaint.textureGenPipeline import Hunyuan3DPaintConfig, Hunyuan3DPaintPipeline
config = Hunyuan3DPaintConfig(max_num_view=6, resolution=512)
config.multiview_pretrained_path = PAINT_MODEL_ID
config.multiview_cfg_path = "hy3dpaint/cfgs/hunyuan-paint-pbr.yaml"
config.custom_pipeline = "hy3dpaint/hunyuanpaintpbr"
realesrgan_ckpt = Path("hy3dpaint/ckpt/RealESRGAN_x4plus.pth")
if realesrgan_ckpt.exists():
config.realesrgan_ckpt_path = str(realesrgan_ckpt)
_paint_pipeline = Hunyuan3DPaintPipeline(config)
_paint_pipeline_name = "hunyuan3d_2_1_paint"
return _paint_pipeline, _paint_pipeline_name
except Exception as paint_import_error:
from hy3dgen.texgen import Hunyuan3DPaintPipeline as LegacyPaintPipeline
try:
_paint_pipeline = LegacyPaintPipeline.from_pretrained(PAINT_MODEL_ID)
_paint_pipeline_name = "hy3dgen_texgen_2_1"
except Exception:
_paint_pipeline = LegacyPaintPipeline.from_pretrained("tencent/Hunyuan3D-2")
_paint_pipeline_name = (
f"hy3dgen_texgen_2_0_fallback({paint_import_error.__class__.__name__})"
)
return _paint_pipeline, _paint_pipeline_name
def _save_uploaded_images(images: Sequence[str], job_dir: Path) -> List[Path]:
if len(images) < 4 or len(images) > 6:
raise gr.Error("Provide 4 to 6 orthographic images: Front, Back, Left, Right, optional Top/Bottom.")
view_names = ["front", "back", "left", "right", "top", "bottom"]
saved_paths: List[Path] = []
for index, source in enumerate(images):
src_path = Path(str(source))
if not src_path.exists():
raise gr.Error(f"Input image not found: {src_path}")
suffix = src_path.suffix.lower() or ".png"
dst_path = job_dir / f"{view_names[index]}{suffix}"
shutil.copy2(src_path, dst_path)
saved_paths.append(dst_path)
return saved_paths
def _prepare_multiview_inputs(image_paths: Sequence[Path]) -> Dict[str, str]:
slot_names = ["front", "back", "left", "right", "top", "bottom"]
return {
slot_names[index]: str(path)
for index, path in enumerate(image_paths)
}
def _extract_mesh(candidate: Any) -> trimesh.Trimesh | None:
if isinstance(candidate, trimesh.Trimesh):
return candidate
if isinstance(candidate, trimesh.Scene):
meshes = [mesh for mesh in candidate.geometry.values() if isinstance(mesh, trimesh.Trimesh)]
if not meshes:
return None
return trimesh.util.concatenate(meshes)
if isinstance(candidate, (list, tuple)):
for value in candidate:
mesh = _extract_mesh(value)
if mesh is not None:
return mesh
return None
def _run_shape_stage(saved_images: Sequence[Path], output_shape_glb: Path) -> Path:
shape_pipeline = _load_shape_pipeline()
multiview_inputs = _prepare_multiview_inputs(saved_images)
shape_result = shape_pipeline(
image=multiview_inputs,
num_inference_steps=30,
octree_resolution=380,
num_chunks=20000,
generator=torch.manual_seed(12345),
output_type="trimesh",
)
shape_mesh = _extract_mesh(shape_result)
if shape_mesh is None:
raise RuntimeError("Shape stage did not produce a valid mesh output.")
shape_mesh.export(output_shape_glb)
return output_shape_glb
def _to_glb(source_path: Path, output_glb_path: Path) -> Path:
loaded = trimesh.load(source_path, force="scene")
loaded.export(output_glb_path, file_type="glb")
return output_glb_path
def _run_paint_stage(
input_mesh_path: Path,
reference_image_path: Path,
output_glb_path: Path,
) -> Tuple[Path, str]:
paint_pipeline, paint_pipeline_name = _load_paint_pipeline()
if paint_pipeline_name.startswith("hunyuan3d_2_1"):
painted_output = paint_pipeline(
mesh_path=str(input_mesh_path),
image_path=str(reference_image_path),
output_mesh_path=str(output_glb_path),
use_remesh=False,
save_glb=True,
)
painted_path = Path(str(painted_output)) if painted_output is not None else output_glb_path
if painted_path.suffix.lower() == ".obj":
glb_candidate = painted_path.with_suffix(".glb")
if glb_candidate.exists():
painted_path = glb_candidate
else:
painted_path = _to_glb(painted_path, output_glb_path)
elif painted_path.suffix.lower() != ".glb":
painted_path = _to_glb(painted_path, output_glb_path)
return painted_path, paint_pipeline_name
shape_mesh = trimesh.load(input_mesh_path, force="mesh")
painted_mesh = paint_pipeline(shape_mesh, image=str(reference_image_path))
if isinstance(painted_mesh, (trimesh.Trimesh, trimesh.Scene)):
painted_mesh.export(output_glb_path)
elif hasattr(painted_mesh, "export"):
painted_mesh.export(output_glb_path)
elif isinstance(painted_mesh, str):
painted_path = Path(painted_mesh)
if painted_path.suffix.lower() == ".glb":
shutil.copy2(painted_path, output_glb_path)
else:
_to_glb(painted_path, output_glb_path)
else:
raise RuntimeError("Paint stage returned an unsupported output format.")
return output_glb_path, paint_pipeline_name
def _model_stats(glb_path: Path, paint_pipeline_name: str, view_count: int) -> Dict[str, Any]:
loaded = trimesh.load(glb_path, force="scene")
if isinstance(loaded, trimesh.Scene):
mesh_list = [mesh for mesh in loaded.geometry.values() if isinstance(mesh, trimesh.Trimesh)]
vertices = int(sum(mesh.vertices.shape[0] for mesh in mesh_list))
faces = int(sum(mesh.faces.shape[0] for mesh in mesh_list))
elif isinstance(loaded, trimesh.Trimesh):
vertices = int(loaded.vertices.shape[0])
faces = int(loaded.faces.shape[0])
else:
vertices = 0
faces = 0
file_size_bytes = glb_path.stat().st_size
return {
"vertices": vertices,
"faces": faces,
"file_size_bytes": file_size_bytes,
"file_size_mb": round(file_size_bytes / (1024 * 1024), 3),
"pbr_channels": ["baseColor", "metallicRoughness", "normal"],
"paint_pipeline": paint_pipeline_name,
"view_count": view_count,
}
@spaces.GPU(duration=120)
def generate_3d_pbr(images: List[str]) -> Tuple[str, Dict[str, Any]]:
_cleanup_temp_jobs()
job_dir = TEMP_ROOT / f"job-{uuid.uuid4().hex}"
job_dir.mkdir(parents=True, exist_ok=True)
try:
saved_images = _save_uploaded_images(images, job_dir)
shape_glb = job_dir / "shape_mesh.glb"
final_glb = job_dir / "model_pbr.glb"
_run_shape_stage(saved_images, shape_glb)
painted_glb, paint_pipeline_name = _run_paint_stage(
shape_glb,
saved_images[0],
final_glb,
)
if painted_glb != final_glb and painted_glb.exists():
shutil.copy2(painted_glb, final_glb)
if not final_glb.exists():
raise RuntimeError("Final GLB file was not generated.")
_register_temp_job(job_dir)
stats = _model_stats(final_glb, paint_pipeline_name, len(saved_images))
return str(final_glb), stats
except Exception as exc:
shutil.rmtree(job_dir, ignore_errors=True)
raise gr.Error(f"Generation failed: {exc}") from exc
with gr.Blocks(title="TwoTh - Hunyuan3D PBR Generator") as demo:
gr.Markdown(
"""
# TwoTh - Serverless 3D PBR Generation
Upload 4 to 6 orthographic views in this order: Front, Back, Left, Right, optional Top/Bottom.
The pipeline runs Hunyuan3D-2mv for mesh generation, then Hunyuan3D-2.1 Paint for PBR texturing.
"""
)
image_inputs = gr.Files(
label="Orthographic Images (4-6)",
file_count="multiple",
file_types=["image"],
)
run_btn = gr.Button("Generate 3D PBR")
model_output = gr.File(
label="PBR GLB Output",
file_types=[".glb"],
)
stats_output = gr.JSON(label="Generation Stats")
run_btn.click(
fn=generate_3d_pbr,
inputs=[image_inputs],
outputs=[model_output, stats_output],
api_name="generate_3d_pbr",
)
demo.queue(default_concurrency_limit=1, max_size=16)
if __name__ == "__main__":
demo.launch()
|