Skip to content
Images

Transparent LoRA для MOSS-Image

RGBA-картинки без вырезания фона

2 min read

genai-craft выпустил LoRA для MOSS-Image-SFT, которая рисует вырезанные объекты, логотипы, надписи и эффекты на прозрачном фоне. Сэмплер сразу отдаёт RGBA PNG, без отдельного удаления фона. Работает вместе с RGBA VAE trmz/flux2-klein-alpha, который декодирует четыре канала; размер LoRA — 149 MB, rank 32.

Ключевые особенности

  • Триггер: в конец промпта добавить isolated on a transparent background; для JSON layout — в summary caption.
  • Rank 32 по q, k, v, o, ffn_up_gate и ffn_down; 504 тензора, fp16, 149 MB.
  • Рецепт: MOSS-Image-SFT в bf16 + 8-step distillation, LoRA применяются unmerged и считаются в fp32; flow-matching, shift 5, 8 шагов.
  • Декод через RGBA VAE с собственными BatchNorm-статистиками и 2x2 unpatchify; выше ~1.2 MP — тайлами.
  • Пример умещает text encoder и transformer на одной 20 GB карте.
  • Лицензия Apache-2.0; обучение на 209 RGBA-картинках, сгенерированных самой MOSS-Image.

Пример

pip install "torch>=2.9" "diffusers>=0.38" "transformers>=5.3" safetensors huggingface_hub pillow
python inference_example.py "photo of a red fox sitting, with a big fluffy tail" fox.png --seed 21
ПодробнееБенчмарк: 0/24 потеряли объект, 0/24 сохранили фон; доля полностью непрозрачных пикселей — 0.39.
Железо: достаточно одной 20 GB карты; на RTX 3080 20GB — 14.9–16.1 с на 1024×1024, на RTX A6000 — около 10.5 с.
Ограничения: заливные карточки и круглые логотипы выходят прозрачными; у дыма и огня жёсткие края, у неона слабое свечение.
Тестировали только с этим рецептом; другие сэмплеры, особенно data-space сэмплер дистилляции, теряют alpha.
Discuss 5

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.

    # genai-craft/transparent-lora-for-moss-image
    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 genai-craft/transparent-lora-for-moss-image --local-dir ~/models/transparent-lora-for-moss-image
    # The model is in ~/models. How to run it is in the model card (link above).
  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 transparent-lora-for-moss-image (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

genai-craft/transparent-lora-for-moss-image
Author
genai-craft
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

05

Ask AI about this model

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

Comments