Skip to content
Video

AURA-1.3B

Audio-Driven Streaming Talking Head Avatar

2 min read

This model from the Kyutai lab turns a portrait image and an audio speech stream into a real-time talking-head video. It is built on a Wan2.1-T2V-1.3B architecture combined with Mimi audio modules (audio tokens) and distilled from a teacher model using DMD (Distribution Matching Distillation). The model outputs 576p video at around 20.7 frames per second on a single H100 GPU while strictly preserving the original face identity (CSIM). This release is intended for researchers and developers building interactive video avatars (under a CC BY-NC 4.0 license).

Key features

  • Real-time streaming generation: outputs results chunk by chunk using a rolling KV cache.
  • High face fidelity: achieves superior identity preservation (CSIM) compared to evaluated methods on the HALLO3 and CelebV-HQ benchmarks.
  • Efficient distillation: a four-step student pipeline (4 denoising steps) accelerates generation by 23x compared to the base teacher model.
  • Two weight formats: standalone files (*_merged) for quick deployment and base checkpoints (LoRA-only) for fine-tuning.

What you can do

  • Animate portraits: generate realistic talking-head videos from a static photo and an audio speech recording.
  • Create an interactive avatar: integrate the model into streaming systems or chatbots to generate video on the fly.
  • Conduct research: use LoRA-only checkpoints and configuration files to customize distillation pipelines (ODE and DMD).

Example

git clone [https://github.com/kyutai-labs/aura](https://github.com/kyutai-labs/aura) && cd aura
uv sync
uv run generate.py --image portrait.jpg --audio speech.wav --output out.mp4
Read moreDetails
HALLO3 benchmarks: FID 56.8, FVD 177.1, CSIM 0.84
Hardware requirements: one H100 GPU cited in the paper (~16-20+ fps)
Architecture: Block-wise causal over chunks of 4 latent frames (~0.64 s)
Limitations: CC BY-NC 4.0 license (non-commercial use only)
Video examples
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 5.9 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 3.0 GB6.4 GB
  • 4-bitweights 1.7 GB4.9 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 5060 Ti 8GB · BF16 — from RTX 3080

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.

    # Mr2cool/AURA-1.3B
    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 Mr2cool/AURA-1.3B --local-dir ~/models/AURA-1.3B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to AURA-1.3B
  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.

Model family

  1. Wan2.1-T2V-1.3Bfamily root
  2. AURA-1.3B-Teacher-TalkVerse
  3. AURA-1.3B

New posts about versions of Wan2.1-T2V-1.3B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 9All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

Mr2cool/AURA-1.3B
Author
Mr2cool
Category
Video
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