Skip to content
Video

Pika H3 Flex Hybrid

адаптер под гибридные чекпоинты MiniMax H3

1 min read

Это блочно собранный LoRA-адаптер для ComfyUI от hyukudan. Он повторяет сплит популярных гибридных чекпоинтов MiniMax H3: блоки 0–29 и все неблочные слои берутся из FL2VA, блоки 30–49 — из Ref2VA. Адаптер не обучался заново: тензоры просто скопированы из двух оригинальных Pika H3 Flex-адаптеров.

Ключевые особенности

  • Подходит только для гибридных чекпоинтов с тем же сплитом: блоки 30–49 из Ref2VA.
  • Блоки 0–29 и неблочные слои — из FL2VA-адаптера, блоки 30–49 — из Ref2VA.
  • Сохраняет мульти-референсное кондиционирование Ref2VA и снижает часть его артефактов.
  • Для чистых FL2VA или Ref2VA моделей нужно использовать оригинальные адаптеры Pika.
  • Style LoRA можно ставить перед этим адаптером в цепочке.

Пример

UNETLoader → minimax_h3_hybrid_fl2va_ref2va_b30-49-int8
LoraLoaderModelOnly → этот адаптер, strength 1.0
Sampler: lcm; BasicGuider, CFG 1.0
ManualSigmas: 1.0, 0.972973, 0.923077, 0.8, 0.571429, 0.333333, 0.0
Без turbo / Lightning LoRA поверх
ПодробнееТестировали на одной GPU 96 ГБ, 1184×672, 24 fps, с image + audio references.
15-секундный кадр с испанским диалогом: чисто, без лишних склеек, стабильная рамка и корректный lip-sync.
25-секундный two-shot с жёсткой склейкой: склейка попала в нужное действие, комната и одежда консистентны.
Скорость: около 28 с/шаг для 15 с и 65 с/шаг для 25 с; это не быстрее 6-step turbo, выигрыш в качестве. Статус — экспериментальный.
Discuss Telegram 3 372

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.

    # hyukudan/Pika-H3-Flex-Hybrid-b30-49
    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 hyukudan/Pika-H3-Flex-Hybrid-b30-49 --local-dir ~/models/Pika-H3-Flex-Hybrid-b30-49
    # 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.
  • 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.

MiniMax H3 prompts

All prompts

Ready prompts with the resulting video: copy and try.

Model family

New posts about versions of Pika-H3-Flex-Hybrid-b30-49 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

hyukudan/Pika-H3-Flex-Hybrid-b30-49
Author
hyukudan
Category
Video
Released
Posted

On NeyroLearn

Hugging Face

3723

Ask AI about this model

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

Comments