Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
"""Rewrite TRELLIS.2's pipeline.json so every weight loads from the staged tree.
Pure stdlib, no torch, so it unit tests anywhere (test_pipeline_config.py).
Upstream's pipeline.json reaches out to three places we do not want at serve
time, and one of them is a licence problem:
* `sparse_structure_decoder` names a repo path in microsoft/TRELLIS-image-large
(MIT). We stage that one checkpoint inside the main tree, so the name is
rewritten to the staged copy.
* `image_cond_model` names the DINOv3 conditioner by Hub id. It is loaded from
the staged copy instead, so the load never touches the Hub. (DINOv3 is
gated on its original repo; the staged bytes come from an ungated mirror
and ship with the DINOv3 License text, see README.md.)
* `rembg_model` names briaai/RMBG-2.0, which is licensed CC BY-NC and cannot
ship in a commercial product. It is replaced with ZhengPeng7/BiRefNet (MIT),
which the same BiRefNet wrapper class loads unchanged.
"""
from __future__ import annotations
import json
import os
from pathlib import Path
from typing import Iterable
# Sub-trees of the staged weights, written by stage_weights.sh.
SS_DECODER_DIR = "ss_dec/ckpts/ss_dec_conv3d_16l8_fp16"
DINOV3_DIR = "dinov3"
BIREFNET_DIR = "birefnet"
REQUIRED_PATHS = (
"pipeline.json",
"ckpts/ss_flow_img_dit_1_3B_64_bf16.safetensors",
"ckpts/shape_dec_next_dc_f16c32_fp16.safetensors",
"ckpts/slat_flow_img2shape_dit_1_3B_512_bf16.safetensors",
"ckpts/slat_flow_img2shape_dit_1_3B_1024_bf16.safetensors",
"ckpts/tex_dec_next_dc_f16c32_fp16.safetensors",
"ckpts/slat_flow_imgshape2tex_dit_1_3B_512_bf16.safetensors",
"ckpts/slat_flow_imgshape2tex_dit_1_3B_1024_bf16.safetensors",
SS_DECODER_DIR + ".safetensors",
DINOV3_DIR + "/model.safetensors",
DINOV3_DIR + "/config.json",
BIREFNET_DIR + "/model.safetensors",
BIREFNET_DIR + "/config.json",
)
def missing_weights(present: Iterable[str]) -> list[str]:
"""Every file the pipeline needs that is not in `present` (relative names)."""
have = set(present)
return [rel for rel in REQUIRED_PATHS if rel not in have]
def localize(config: dict, aux_root: str | os.PathLike[str]) -> dict:
"""Return a copy of `config` that loads the conditioner and matte locally.
`aux_root` holds the dinov3/ and birefnet/ directories, which are small
enough (about 1.7 GB together) to keep on local disk for the instance's life.
"""
base = str(Path(aux_root))
out = json.loads(json.dumps(config))
args = out["args"]
args["models"]["sparse_structure_decoder"] = SS_DECODER_DIR
args["image_cond_model"]["args"]["model_name"] = f"{base}/{DINOV3_DIR}"
args["rembg_model"] = {
"name": "BiRefNet",
"args": {"model_name": f"{base}/{BIREFNET_DIR}"},
}
return out
def write_local_config(
pipeline_json: str, aux_root: str | os.PathLike[str], out_dir: str | os.PathLike[str]
) -> Path:
"""Write the localized pipeline.json into `out_dir` and return its path."""
target = Path(out_dir)
target.mkdir(parents=True, exist_ok=True)
path = target / "pipeline.json"
path.write_text(json.dumps(localize(json.loads(pipeline_json), aux_root), indent=2))
return path
def link_weight_tree(root: str | os.PathLike[str], out_dir: str | os.PathLike[str]) -> None:
"""Symlink each top-level entry of a local weight tree into `out_dir`.
The pipeline resolves every model as `<config dir>/<relative name>`, so a
config dir holding the localized pipeline.json plus these links loads
straight from a mounted or pre-staged tree.
"""
source = Path(root)
target = Path(out_dir)
target.mkdir(parents=True, exist_ok=True)
for entry in source.iterdir():
if entry.name == "pipeline.json":
continue
link = target / entry.name
if not link.exists() and not link.is_symlink():
link.symlink_to(entry.resolve())