Skip to content
Video

ReDiverse-Pika-H3-Flex

bringing diversity back to video generation

2 min readupdated

Fast video generation models often suffer from a common drawback: changing the seed (initial noise) produces nearly identical clips. The ReDiverse-FT adapter from researchers at UIUC, Johns Hopkins University, and fal fixes this in Pika-H3-Flex—a distilled 6-step version of MiniMax-H3 capable of generating video with synchronized audio. It is a lightweight LoRA add-on (rank 32, 612 MB) that only activates during the very first generation step, leaving the model to finish the remaining five steps as usual. As a result, rendering stays just as fast, while different seeds actually produce distinct scenes.

Key features

  • Works on top of the MiniMax-H3 base model paired with the Pika-H3-Flex LoRA adapter (text-to-video mode).
  • Trained to match the activations of the original teacher model across blocks up to transformer_blocks.45.
  • Activates only on the first sampling step out of six, leaving the rest of the generation speed untouched.
  • Supports video generation with synchronized audio, defaulting to 768p resolution (16:9, 5 seconds) with an option to switch to 480p.

What you can do

  • Generate noticeably different video variations from the exact same text prompt simply by cycling through seed values.
  • Run a side-by-side comparison between standard Pika-H3-Flex output and the adapter version using the same noise via the built-in --compare flag.
  • Pair it with Pika-H3-Flex to rapidly prototype 5-second audio-synced clips when you need varied takes.

Example

hf download ReDiverse/ReDiverse-Pika-Flex --local-dir ReDiverse-Pika-Flex
python ReDiverse-Pika-Flex/generate_pika.py --prompt "a corgi running on the beach at sunset" --seeds 0-3 --out out --compare
Read moreDetails
The rediverse_ft.safetensors file weighs 612 MB (rank 32) and stacks on top of the official Pika LoRA.
Before running it, you need to install the Pika H3 Flex package, which automatically downloads the base MiniMax-H3 weights.
Trained and tested exclusively for text-to-video tasks.
License: MiniMax H3 Community License Agreement (research only; regional restrictions apply, including bans in the EU, US, UK, and South Korea).
Discuss Telegram 4 20

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 62 GB · ready files75 GB
  • FP8 / INT8recommendedweights 31 GB39 GB
  • 4-bitweights 18 GB24 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.

This is a LoRA: counted together with its base model Pika-H3-Flex.

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.

    # ReDiverse/ReDiverse-Pika-H3-Flex
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated 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 Pika-Labs/Pika-H3-Flex --local-dir ~/models/Pika-H3-Flex
    hf download ReDiverse/ReDiverse-Pika-H3-Flex --local-dir ~/models/ReDiverse-Pika-H3-Flex
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Pika-H3-Flex
  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

Requires the base model

Pika-Labs/Pika-H3-Flex

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. MiniMax-H3family root
  2. Pika-H3-Flex
  3. ReDiverse-Pika-H3-Flex

New posts about versions of MiniMax-H3 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 3All on Hugging Face

About the model

ReDiverse/ReDiverse-Pika-H3-Flex
Author
ReDiverse
Category
Video
Released
Posted

On NeyroLearn

Hugging Face

204

Ask AI about this model

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

Comments