Skip to content
3D

Idol Turbo

3D-гауссианы из одной картинки за 5 шагов

2 min readupdated

Idol Turbo от akhaliq — дистиллированная few-step версия speridlabs/idol для image-to-3D. На входе одно фото объекта, на выходе до миллиона 3D-гауссианов: 5 шагов без guidance вместо 50 шагов с classifier-free guidance.

Ключевые особенности

  • 100 network evaluations → 5, сокращение NFE ~20×; ~0,75 с на H200 (0,66 с при 4 шагах, ~1,0 с при 8).
  • Glance-style phase-LoRA: два rank-32 эксперта — Slow-LoRA для высокого шума и Fast-LoRA для низкого, переключение на t=0.5; guidance вшит в веса, 29,4M обучаемых параметров, база не тронута.
  • v5: 4 650 записей учителя (50 шагов, CFG 3.0) на 93 разных изображениях объектов, 2 000 шагов обучения (~25 мин на одной H200).
  • Совместим с Comfy-Org/Idol и TripoSplat workflows: LoRA-ключи совпадают, strength 1.0; настройки — 5 шагов, CFG 1.0.
  • Для полной точности можно связать две стадии сэмплера: slow на первые ~4 из 5 шагов, fast на финиш; простой путь — только slow-LoRA.

Пример

Нужен idol SDK и доступ к gated-энкодерам.

from idol import IdolPipeline
from huggingface_hub import hf_hub_download
import importlib.util
spec = importlib.util.spec_from_file_location("turbo", hf_hub_download("akhaliq/idol-turbo", "turbo.py"))
turbo = importlib.util.module_from_spec(spec); spec.loader.exec_module(turbo)
pipeline = IdolPipeline.from_pretrained("akhaliq/idol-turbo")
turbo.patch_pipeline_for_turbo(pipeline, "akhaliq/idol-turbo")
output = pipeline("photo.jpg", mode="real", seed=42, steps=5, cfg_strength=0.0)
ПодробнееБенчмарки: ~0,75 с на H200 при 5 шагах; 0,66 с при 4 и ~1,0 с при 8; ~20× меньше NFE.
Обучение: 4 650 траекторий, 93 изображения, 2 000 шагов, ~25 мин на H200.
Ограничения: часть входов мягче 50-шагового учителя; на 4 шагах цвета могут уходить в тёплый; насыщенные объекты чуть менее яркие, чем в раннем чекпоинте на 10 изображениях.
ComfyUI: двухстадийный сплит не проверяли внутри самого ComfyUI; безопасный дефолт — slow-only.
База speridlabs/idol на 50 шагах — эталон качества. Лицензия Apache-2.0.
Примеры и видео
Discuss Telegram 38 391

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # akhaliq/idol-turbo
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download akhaliq/idol-turbo --local-dir ~/models/idol-turbo
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to idol-turbo
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of idol-turbo (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

akhaliq/idol-turbo
Author
akhaliq
Category
3D
Released
Posted

On NeyroLearn

Hugging Face

39138

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

3D

Idol

Idol