Skip to content
Images

Sana-Sprint CoreML

Converting a Fast Image Generation Model for Apple Devices

2 min readupdated

This repository contains an unofficial conversion of the Sana-Sprint 1.6B image generation model (originally developed by NVIDIA and MIT Han Lab) into the Core ML format for local execution on iPhones and other Apple devices using the Neural Engine (a specialized AI chip). The model generates 1024x1024 pixel images in just 1, 2, or 4 steps using the TrigFlow sampling method. The project is aimed at developers and enthusiasts who want to run fast diffusion generation directly on mobile hardware.

Key features

  • Core ML optimization: The architecture is split into components for efficient execution on the Apple Neural Engine and CPU.
  • Ultra-fast generation: 1024x1024 image creation in 1–4 steps thanks to modern sampling approaches.
  • 8-bit weights: The model and text encoder use 8-bit quantization to save on-device memory.
  • Built-in encoder: The Gemma-2-2B-based text encoder comes as prepared chunks with a pre-installed fixed prompt instruction.

What you can do

  • iOS app integration: Add local text-to-image generation to your own iOS mobile application.
  • Hardware experiments: Test the inference speed of next-generation diffusion transformers on Apple Silicon processors and the Neural Engine.
  • On-device generation: Create detailed 1024x1024 pixel images completely offline without sending data to a server.

Example

The model comes as compiled Core ML archives (`.mlmodelc.aar`) including the text encoder, the diffusion transformer with linear attention, and the DC-AE autoencoder for image decoding.

Read moreDetails
Image resolution: 1024x1024 pixels
Generation steps: 1, 2, or 4 steps (TrigFlow)
Architecture: SANA-Sprint 1.6B + Gemma-2-2B (text encoder with 8-bit weights)
License: Apache License 2.0 (model code), text components are subject to the Gemma Terms of Use
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.5 GB · ready files2.1 GB
  • FP8 / INT8recommendedweights 0.3 GB1.8 GB
  • 4-bitweights 0.2 GB1.7 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.

    # 33cloud33/Sana-Sprint-CoreML
    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 33cloud33/Sana-Sprint-CoreML --local-dir ~/models/Sana-Sprint-CoreML
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Sana-Sprint-CoreML
  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

  1. Sana_Sprint_1.6B_1024pxfamily root
  2. Sana_Sprint_1.6B_1024px_diffusers
  3. Sana-Sprint-CoreML

New posts about versions of Sana_Sprint_1.6B_1024px (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 5All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

33cloud33/Sana-Sprint-CoreML
Author
33cloud33
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