Skip to content
Video

Sumi-e LoRA for LTX-2.5

Japanese ink wash painting style videos

2 min readupdated

This is a style LoRA (a lightweight adapter on top of a base model) for the Lightricks LTX-2.5 video generator by enoky. It gives generated videos the look of traditional Japanese sumi-e ink wash painting, complete with black ink brushstrokes, gray washes, rice paper texture, and generous negative space. Because the adapter was trained strictly on static frames, it controls only the visual style while leaving motion smoothness entirely to the base LTX-2.5 model. It is ideal for creators looking for artistic animation in a minimalist Eastern aesthetic.

Key features

  • Trigger word activation: Start your prompt with sumi-e ink painting, to apply the style.
  • Compatibility: The .safetensors weights (rank 32) work out of the box with ComfyUI and standard ltx-pipelines.
  • Preserves base motion: Trained on single-frame clips targeting only the video branch (self-attention, text cross-attention, feed-forward), so physics and movement remain native to LTX-2.5.
  • Close-up bias: Since the dataset consisted of close-ups, the model performs best when you explicitly specify framing in your prompt.

What you can do

  • Generate minimalist artistic intros and aesthetic clips for social media or video essays with ink bleeding across paper.
  • Create animated close-up nature studies, such as bamboo leaves or tree branches swaying in the wind.
  • Pair the adapter with the base ltx-2.5-22b-dev transformer in ComfyUI, dialing the strength between 0.7 and 1.0.

Example

uv run python -m ltx_pipelines.ti2vid_one_stage \
    --transformer-path models/ltx-2.5/diffusion_models/ltx-2.5-22b-dev-transformer-bf16.safetensors \
    --lora sumi-e_ltx-2.5_lora.safetensors 1.0 \
    --prompt "sumi-e ink painting, Close-up of bamboo leaves swaying in the wind, wet ink bleeding into rice paper." \
    --output-path out.mp4
Read moreDetails
Training base: ltx-2.5-22b-dev-transformer-bf16, 2000 steps (rank 32) via ltx-trainer.
Dataset: curated subset of synthetic images from takara-ai/Sumi-e_no_kurozuappu_ink_painting_closeups.
Limitations: strong bias toward close-up framing and potential inheritance of artifacts from the generator used to build the dataset.
License: LTX-2.x Community License (derivative of LTX-2.5).
Discuss Telegram 0

Will it run?

Only with RAM offload

Even 4-bit is larger than your VRAM. Some layers can stay in system RAM (llama.cpp, ComfyUI), but it will be several times slower.

  • BF16weights 74 GB · ready files88 GB
  • FP8 / INT8recommendedweights 37 GB45 GB
  • 4-bitweights 21 GB27 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: Won't fit any home GPU — renting is cheaper.

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

This is a LoRA: counted together with its base model LTX-2.5.

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.

    # enoky/sumi-e-ltx-2.5-lora
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Lightricks/LTX-2.5 --local-dir ~/models/LTX-2.5
    hf download enoky/sumi-e-ltx-2.5-lora --local-dir ~/models/sumi-e-ltx-2.5-lora
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LTX-2.5
  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.
  • LightX2V step distill0.6.1from November 9, 2026new release

    5–10× fewer steps

    Distilled LoRAs and an engine: video in 4–8 steps instead of 30–50.

    How to enable: Load their step-distill LoRA, CFG 1.0, 4–8 steps.

  • 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

Requires the base model

Lightricks/LTX-2.5

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of LTX-2.5 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 3All on Hugging Face

About the model

enoky/sumi-e-ltx-2.5-lora
Author
enoky
Category
Video
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