Pixal3D model set pixal3d-sv-q8_0 v1 (single-view)

GGUF weights for pixal3d.cpp, a GGML-based C++ image→3D pipeline. This is the single-view (SV) Web release set: Q8_0 quantization, 9 files, 7.54 GiB. It is the one-image counterpart of the multiview set raven38/pixal3d-q8_0-v1: the four flow DiTs are the official Pixal3D single-view checkpoints, and the five shared models (image encoder, NAF, three decoders) are byte-identical to the multiview set.

The repository layout is flat and the file names are exact, because the consuming client verifies every file's size_bytes and SHA-256 against the committed manifest pixal3d-models.json (model_family: sv) before the set is marked usable. The Web app never substitutes multiview weights for a single image and never substitutes these weights for a multiview run; the two families are selected only through their own manifests.

Files

Provenance

The four *_sv.gguf files were converted with tools/convert.py from the official TencentARC/Pixal3D single-view checkpoints ckpts/ss_flow_img_dit_1_3B_64_bf16.safetensors, ckpts/slat_flow_img2shape_dit_1_3B_512_bf16.safetensors, ckpts/slat_flow_img2shape_dit_1_3B_1024_bf16.safetensors and ckpts/slat_flow_imgshape2tex_dit_1_3B_1024_bf16.safetensors (the variants without the _mv suffix) to F16 GGUF, then quantized with tools/quantize_gguf.py <src> <dst> Q8_0. The SV and MV checkpoints share the same config and tensor layout; only the trained values differ.

Usage

Point the Pixal3D Web app at this repository as the single-view model source:

sv_manifest_url   = https://huggingface.co/raven38/pixal3d-sv-q8_0-v1/resolve/main/pixal3d-models.json
sv_model_base_url = https://huggingface.co/raven38/pixal3d-sv-q8_0-v1/resolve/main

(query parameters ?sv_manifest_url=…&sv_model_base_url=…, or the deployment's injected PIXAL3D_SV_MODEL_* values). For the native CLI, place the nine files in one directory and run trellis-cli with --pixal3d-weights sv and a single --views entry. The client streams each file into origin-owned storage and refuses the set unless every file matches the manifest exactly.

Licensing — read before redistributing

This repository contains files under two different licenses. LICENSE.md is the per-file index; the two license texts are MIT_LICENSE.md and DINOV3_LICENSE.md.

dinov3.gguf — DINOv3 License (not MIT)

dinov3.gguf is a format conversion of Meta's DINOv3 ViT-L/16 (lvd1689m) and remains governed by the DINOv3 License, a copy of which is included in this repository as DINOV3_LICENSE.md.

The DINOv3 License grants the right to "use, reproduce, distribute, copy, create derivative works of, and make modifications to the DINO Materials" (§1.a), and this conversion is distributed under that grant. If you redistribute dinov3.gguf or any derivative of it, §1.b.i requires that you do so only under the terms of that same Agreement and that you provide a copy of the Agreement alongside it. §1.b.ii requires acknowledging DINO Materials in research publications. §1.b.iii and §1.b.v impose Trade Control / ITAR restrictions and prohibit military, nuclear, espionage and weapons end uses. §1.b.iv prohibits reverse engineering the materials.

All other files — MIT

The remaining files derive from the TRELLIS.2 backbone (MIT, © Microsoft Corporation) and from Pixal3D training on top of it (MIT, © 2026 Tencent). Both copyright notices and the MIT permission notice are reproduced in MIT_LICENSE.md, which you must keep when redistributing them. The pixal3d.cpp runtime is MIT as well.

Contents of the weights

The GGUF metadata in every file consists of general.architecture, general.name and, for eight of the nine, a trellis.config_json holding model hyperparameters. No file paths, usernames, hostnames, timestamps or contact information are embedded.