Skip to content
Images

NAMI-T2I

эффективная генерация изображений на 2B параметров

2 min readupdated

Модель NAMI-T2I (NAMI-2B) от команды 360ZhiNao (360CVGroup) решает задачу генерации картинок по текстовому описанию (text-to-image). В основе лежит архитектура Bridged Progressive Rectified Flow Transformers (подход на базе выпрямленных потоков для ускоренного и качественного синтеза), а научная работа по проекту принята на конференцию CVPR 2026. Для обработки текста система опирается на энкодеры mCLIP и mT5-XXL. Подойдёт исследователям и разработчикам, которые ищут компактный генератор изображений на 2 миллиарда параметров с открытыми весами для тестирования и деплоя.

Ключевые особенности

  • Компактный размер: опубликованный чекпоинт NAMI-2B содержит 2 миллиарда параметров и предназначен для эффективного инференса.
  • Два текстовых энкодера: в связке с моделью используются токенизаторы и энкодеры mclip (от HunyuanDiT) и mt5-xxl (от Google).
  • Академическая база: архитектура Bridged Progressive Rectified Flow Transformers прошла рецензирование и принята на CVPR 2026.
  • Готовность к развёртыванию: авторы выложили веса и код инференса для локального тестирования и деплоя.

Что можно сделать

  • Развернуть локальный пайплайн генерации изображений по тексту с использованием кода из репозитория 360CVGroup/NAMI-T2I и опубликованных весов.
  • Протестировать качество понимания сложных промптов за счёт комбинации энкодеров mCLIP и mT5-XXL.
  • Сравнить скорость и качество генерации Rectified Flow-подхода NAMI-2B с другими моделями класса 2B.

Пример

Для запуска генерации необходимо скачать веса NAMI-2B в директорию weights, а также подгрузить токенизаторы и текстовые энкодеры из репозиториев Tencent-Hunyuan/HunyuanDiT (папка t2i/clip_text_encoder) и google/mt5-xxl, после чего запустить инференс через официальный код проекта 360CVGroup/NAMI-T2I.

ПодробнееАрхитектура: Bridged Progressive Rectified Flow Transformers (2B параметров)
Текстовые энкодеры: mCLIP (HunyuanDiT) и mT5-XXL
Публикация: статья принята на CVPR 2026 (arXiv:2503.09242)
Примечание: в самой карточке на Hugging Face бенчмарки и требования к VRAM не указаны.
Discuss Telegram 1 10

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.0 GB · ready files10 GB
  • FP8 / INT8recommendedweights 4.0 GB5.9 GB
  • 4-bitweights 2.3 GB4.0 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 2080 Ti

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.

    # 360ZhiNao/NAMI-T2I
    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 360ZhiNao/NAMI-T2I --local-dir ~/models/NAMI-T2I
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to NAMI-T2I
  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 NAMI-T2I (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

360ZhiNao/NAMI-T2I
Author
360ZhiNao
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

101

Ask AI about this model

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

Comments