HunyuanVideo Enhance A Video

Most acceleration and quality nodes in this pack ask you to trade something for something. This one is the exception, and kijai says so directly in the README: Enhance-A-Video (the technique this node wraps, from NUS-HPC-AI-Lab) has "a very slight hit on inference speed and zero hit on memory use," and his own initial tests indicated it's "absolutely worth using." That's about as close to an unqualified recommendation as this changelog gets - worth paying attention to.

The mechanism, per the linked project, is a cross-frame intensity adjustment inside the attention layers that pushes video diffusion models toward more coherent, higher-quality motion without retraining anything. You're not adding a LoRA or changing the model - this is a sampling-time nudge, applied through a settings node that plugs into HyVideoSampler's optional feta_args input.

The inputs that matter

• weight (default 2) - the strength of the effect. This is the one dial worth touching; start at default, and if you want to push it, do so gradually rather than jumping to a large value.
• single_blocks / double_blocks (both default on) - which of the transformer's two block types the effect applies to. Leave both on unless you have a specific reason to isolate one.
• start_percent / end_percent (default 0 to 1) - applies across the whole denoising schedule by default. Narrowing this is an option if you only want the enhancement during, say, the later detail-refining steps rather than early structure formation.

Output: feta_args, straight to the sampler.

How to install it

Ships with the pack - nothing to download separately for this node.

• ComfyUI Manager - search ComfyUI-HunyuanVideoWrapper, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-HunyuanVideoWrapper, then pip install -r ComfyUI-HunyuanVideoWrapper/requirements.txt, restart.

No model files needed for this node specifically - it's a pure sampling-time technique layered on top of whatever model you've already got loaded.

Common issues & troubleshooting

There isn't much to go wrong here, which is itself worth saying plainly. If you connect this node and see no change at all, check that feta_args is actually wired into HyVideoSampler - a disconnected settings node is the most common reason any of these nodes appear to do nothing.

If you push weight well past the default and start seeing oversaturated or artifact-heavy motion, that's the effect being applied too strongly - back it down toward 2 rather than assuming something's broken. Beyond that, this is one of the lower-risk nodes in the whole pack: given the near-zero memory cost and minimal speed cost the README describes, there's little downside to just leaving it connected at default settings on any HunyuanVideo generation you run through this wrapper.