Skip to content
Images

LaRenderer: A 3D Scene Neural Renderer Supporting FLUX.2

2 min readupdated

LaRenderer transforms 3D scenes featuring PBR materials, glass, mirrors, and point lights into high-quality images from any camera angle. Its architecture encodes scenes into sparse voxels (using TRELLIS.2 SC-VAE) and generates latent representations through a flow-matching decoder, followed by sRGB decoding. The model is released in both full teacher weights (1.071B parameters) and a distilled single-step student (0.722B parameters), supporting four operating modes. This tool is designed for researchers and developers working on neural rendering and the generation of synthetic 3D environments.

Key features

  • Four inference modes: Flexible choice between 5-step generation (including pair-sync variants) and ultra-fast single-step decoding.
  • Third-party component integration: The model relies on powerful external networks, including FLUX.2 VAE and TRELLIS.2 SC-VAE.
  • Efficient distillation: The single-step student shares the scene encoder with the teacher for instant generation.
  • Precise handling of scene physics: Support for mirrors, glass, point light sources, and environments via HDR maps.

What you can do

  • Render synthetic rooms: Generate high-quality images of procedurally created scenes around 4 meters in size.
  • Algorithm research: Test 5-step and 1-step modes with pair-sync to evaluate quality on benchmarks.
  • Pipeline integration: Use the weights alongside the neural rendering repository to prepare and render custom scenes.

Example

huggingface-cli download eternity304/LaRenderer --local-dir ./LaRenderer
python scripts/inference/prepare_scene.py --bundle my_scene --out shards/my_scene
python scripts/inference/larender_infer.py --release ./LaRenderer \
    --flux2-vae ./deps/flux2_klein_4b/vae/diffusion_pytorch_model.safetensors \
    --shard shards/my_scene/shard-00000.tar --mode 1step --out renders/my_scene
Read moreDetails
Benchmarks (Caustic benchmark v1, 24 scenes / 72 views, 512px): The 1step_pairsync mode achieves top results (PSNR 23.48, SSIM 0.910, LPIPS 0.121).
Hardware and dependency requirements: External networks (FLUX.2 VAE, TRELLIS.2 SC-VAE, Qwen-Image VAE) are required and must be downloaded separately.
Limitations: The model was trained on rooms approximately 4 meters in size, does not support HDR output (produces display-referred AgX sRGB), and exhibits artifacts when lights are placed very close to surfaces.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.0 GB · ready files5.9 GB
  • FP8 / INT8recommendedweights 2.0 GB3.7 GB
  • 4-bitweights 1.1 GB2.7 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # eternity304/LaRenderer
    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 eternity304/LaRenderer --local-dir ~/models/LaRenderer
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LaRenderer
  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.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • 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 (3)
    • 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.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • 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 LaRenderer (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

eternity304/LaRenderer
Author
eternity304
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments