Workflows/Create a 3D object from an image with Idol

Images / WORKFLOW

Create a 3D object from an image with Idol

Install Speridlabs’ Idol SDK on an NVIDIA Linux machine or a rented GPU, turn one object photo into a 3D Gaussian splat, and rotate the result in a viewer.

Verified sourceSource checked 11/9/2026

LOCALRENTED GPU

Freshness 94 out of 100. Release age: 17 days old, Oct 28, 2026. Measures age, not quality.

Popularity 65 out of 100. 31,367 downloads in 30 days; 2,259 lifetime likes. Checked Nov 13, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

A new release starts at 100. The score halves every 180 days: 100 × 2^(−age in days / 180). When the release date is unavailable, a repository date is a fallback—not proof of when the model was released. Missing dates stay unscored.

Check the date source

Popularity · adoption

Downloads over the past 30 days carry 80% of the score; lifetime likes carry 20%. Both use a logarithmic scale, capped at 10 million downloads and 100,000 likes. If likes are unavailable, downloads carry all the weight. Missing download statistics stay unscored.

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Nov 14, 2026. Statistics checked Nov 13, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

Before you begin

Hardware

Request times are Speridlabs’ published medians for one image at 262,144 Gaussians and 50 steps. Not measured by YouRunAI.

Machine	Idol · 262,144 Gaussians	Idol · 1,048,576 Gaussians
NVIDIA RTX 3080 · 10 GB VRAM	Works; publisher median 46.2 s per image	Likely too tight; use the default setting
NVIDIA RTX 4090 · 24 GB VRAM	Publisher median 19.1 s per image	Expected to fit; slower than default (estimate)
NVIDIA RTX 5090 · 32 GB VRAM	Publisher median 14.1 s per image	Expected to fit (estimate)
RunPod H100 · 80 GB VRAM	Publisher median 6.7 s per image	Comfortable
AMD Radeon · 16 GB VRAM	Not supported: requires an NVIDIA CUDA GPU; use RunPod	Not supported: requires CUDA; use RunPod
Intel Arc · 16 GB VRAM	Not supported: requires an NVIDIA CUDA GPU; use RunPod	Not supported: requires CUDA; use RunPod
Apple Silicon · 32 GB unified memory	Not supported: no CUDA on Mac; use RunPod or the demo Space	Not supported: no CUDA; use RunPod
CPU only · 32 GB RAM	Not supported: requires an NVIDIA CUDA GPU; use RunPod	Not supported: requires CUDA; use RunPod

IMAGE TO 3D · NVIDIASOURCE-LINKED SETUP

Create a 3D object from an image with Idol

Accept two encoder licenses, install the publisher SDK, and turn one photo into a .ply Gaussian splat you can rotate. Runs on an NVIDIA Linux machine or a rented GPU.

Choose a package Run its commands Check the result

BEFORE YOU BEGIN

Choose your package

Select the weights you want. The instructions, downloads and settings change with your choice.

The publisher has not provided a complete memory target for this task. Check its hardware guidance before starting.

CURRENT SETUP

Idol · 262,144 Gaussians · official

Publisher-reported 7.2 GB peak at this setting; their smallest tested card is a 10 GB RTX 3080.

Freshness 94 out of 100. Release age: 17 days old, Oct 28, 2026. Measures age, not quality.

Popularity 65 out of 100. 31,367 downloads in 30 days; 2,259 lifetime likes. Checked Nov 13, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

Check the date source

Popularity · adoption

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Nov 14, 2026. Statistics checked Nov 13, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

View weight source

THE WORKFLOW

Set up Idol · 262,144 Gaussians on your machine

Choose your computer first, then its GPU path. The instructions below update to match your selection.

CHOOSE YOUR SETUPLinux

Prepare Python and PyTorch with CUDA

Idol needs Linux with an NVIDIA GPU and CUDA, Python 3.10 or later and PyTorch 2.5 or later. Create a virtual environment in Terminal and install the CUDA build of PyTorch using the selector on pytorch.org, then confirm that PyTorch sees the GPU.

Install PyTorch with CUDA Idol requirements

Bash

python3 -m venv ~/idol-env
source ~/idol-env/bin/activate
python -c "import torch; print(torch.__version__, torch.cuda.is_available())"

Accept the two gated encoder licenses

Idol reads images through two gated models. Sign in to Hugging Face and accept the terms on both pages below, then create a read token in your Hugging Face settings. Only the VAE from FLUX.2-dev is downloaded.

facebook/dinov3-vith16plus-pretrain-lvd1689m black-forest-labs/FLUX.2-dev

Install the Idol SDK and log in

In Terminal, clone the publisher’s SDK, install it and log in with your Hugging Face token. There is no separate weight download: the first run fetches Idol (about 3 GB) and the encoders into the Hugging Face cache, and stops with a clear message if an encoder license is still missing.

Idol SDK on GitHub

Bash

git clone https://github.com/speridlabs/idol-sdk.git
cd idol-sdk
pip install -e .
hf auth login

Generate a 3D splat from one photo

Copy a photo of a single object (for example a mug or a shoe on a plain table) into the idol-sdk folder as photo.jpg. Set the input to that file and run Idol in real-photo mode with the default 262,144 Gaussians, 50 steps and seed 42. The first image is slower while PyTorch compiles. Use --mode synthetic instead for renders or AI-generated images.

Setting	Value
Input	One object photo, photo.jpg
Mode	real (photographs)
Gaussians	262,144
Steps	50 (default)
Seed	42 (default)

Idol flags and outputs

Bash

python -m idol.sample --image photo.jpg --mode real --out outputs

Or use the publisher’s browser app

If you prefer clicking to typing, the SDK includes the Gradio demo that powers Speridlabs’ Hugging Face Space. Install its extra packages and start it from the idol-sdk folder in Terminal, then open the local address it prints, upload your photo, generate, and download the .ply file. It uses the same model and downloads, so the checks below still apply. Stop it with Ctrl+C.

Gradio demo in the Idol SDK

Bash

pip install -e ".[demo]"
cd demo
python app.py

Check the three output files

Open the outputs folder. photo.ply is the 3D Gaussian splat, photo_input.png is the cropped, background-removed image the model actually used, and photo_camera.json is its predicted camera. Check photo_input.png first: if the crop cut off part of the object, retake the photo with more space around it.

Open the splat in a viewer and rotate it

Drag photo.ply into a Gaussian splat viewer such as SuperSplat in your browser and rotate it to see the sides the photo never showed. The result is a splat (millions of coloured, semi-transparent points), not a mesh: it is for viewing and splat-based scenes, not for 3D printing or rigging without a separate conversion.

SuperSplat browser editor

SUCCESS CHECK · Idol · 262,144 Gaussians

A rotatable 3D splat of your object

outputs/photo.ply opens in a splat viewer and shows a plausible back and sides of the object, and photo_input.png shows the crop the model used.

Idol outputs a 3D Gaussian splat, not a mesh. On your own Linux PC you can use either the command line or the publisher’s Gradio browser app; the RunPod route uses the command line. Community ComfyUI nodes and Apple Silicon forks exist but are not publisher routes, so they are not covered here. Checked November 9, 2026; not run by YouRunAI.

Back to setup steps

When it doesn’t go to plan

Idol stops saying an encoder is not accessible.

Accept the license on both gated pages with the same Hugging Face account, then run hf auth login again with a read token.

torch.cuda.is_available() prints False.

Install the CUDA build of PyTorch from pytorch.org inside the activated environment and check nvidia-smi shows the GPU.

The object comes out cut off or merged with the background.

Use a photo with one object, a plain background and space around it. Check photo_input.png to see the crop Idol used.

REFERENCE LIBRARY

Sources and files

Original instructions, model files, and compatibility notes behind this setup.

3 SOURCES

01Idol model card: requirements, flags and timingshuggingface.co 02Idol SDKgithub.com 03Hosted demo Spacehuggingface.co

The model behind this workflow

Idol

Keep exploring

Text

SETUP GUIDE

Qwen3.5 · 9B private chat setup

Source-backed package and platform-route plan for private local chat with Qwen3.5 · 9B. Preserves the official Ollama 9B package, a smaller 4B family fallback, and a clearly labeled community uncensored GGUF option.

Freshness 37 out of 100. Release age: 260 days old, Feb 27, 2026. Measures age, not quality.

Build a private coding assistant with Qwen3.8 27B

Run Qwen3.8 27B with Ollama, connect it to Cline in VS Code, and make one reviewable code change in a tiny local project.

Freshness 68 out of 100. Repository age: 101 days old, Aug 5, 2026. Measures age, not quality.

Arena Open Source Web coding: rank 13 of 63. Ranked model: Qwen3.8-27B. Original model ranking, not a benchmark of this package. Export dated Nov 11, 2026. Source checked Nov 13, 2026. Human preference ranking, not hardware suitability.+3

SOFTWAREOllama + Visual Studio Code + Cline + Python

Create images locally with FLUX.2 Klein

Generate images from text prompts in ComfyUI with the fast distilled or more adjustable base 4B model. Save your results and reuse the visual workflow.

Freshness 31 out of 100. Release age: 303 days old, Jan 15, 2026. Measures age, not quality.

SOFTWAREComfyUI

LOCALRENTED GPU

Verified source