Skip to content
Images

Big-LaMa for mobile

models for on-device inpainting

2 min readupdated

This is a conversion of the well-known Big-LaMa model for the mobile platforms Core ML (for iOS) and LiteRT (for Android), designed for object removal and background filling. The original model weights remain unchanged, but the formats were adapted to run on smartphones and tablets. The iOS variant is 50 MB and optimized with 8-bit weight compression (palettization), while the Android version uses matrix products instead of Fourier transforms to run on GPUs and NPU accelerators. This release will be useful for mobile app developers who need high-quality background inpainting directly on the device.

Key features

  • Original Big-LaMa weights with no loss in inpainting quality
  • Core ML support for iOS 17+ with 8-bit weight quantization
  • LiteRT (Android) version replacing FFT with matrix operations for NPU execution
  • Optimized inference speed (~250 ms on Mac GPU and ~184 ms on Snapdragon 8 Elite NPU)

What you can do

  • Integrate the inpainting model into an iOS app to remove objects from photos
  • Add background infilling to an Android app using hardware NPU accelerators
  • Use ready-to-use .mlpackage files or compiled graphs in your computer vision projects

Example

The model inputs require an image and a mask of 512×512 pixels. For Core ML, inference is run via computeUnits = .cpuAndGPU, since fast Fourier transforms are not supported directly on the Neural Engine.

Read moreDetails
Benchmarks: mean absolute error of 0.050 for Core ML and 3e-6 for TFLite.
Platforms: iOS 17+ (Core ML) and Android / LiteRT.
License: Apache-2.0.
Discuss Telegram 0 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.

    # Xalcker/perspectia-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 Xalcker/perspectia-models --local-dir ~/models/perspectia-models
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to perspectia-models
  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 perspectia-models (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Xalcker/perspectia-models
Author
Xalcker
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

50

Ask AI about this model

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

Comments