FlashSR Hybrid Crossover – Preserve Original / Blend Air

FlashSR is a real thing and it's genuinely good at one trick: it reconstructs high-frequency audio content that the source never had. Feed it a 32 kHz MiniMax track and it invents the upper spectrum. But "invents" is doing a lot of work in that sentence, and the pure-FlashSR output is where the community complaints about watery, smeary cymbals come from. The FlashSR Hybrid Crossover is this pack's answer: instead of replacing your audio with FlashSR's version, it keeps the cleanly resampled original intact and blends in only a controlled band of FlashSR's reconstructed high end - the "air."

That's the core insight of the whole pack, and it's worth re-stating because it's the opposite of how most people first wire up super-resolution. You don't trust the enhancer with your whole signal. You trust it only with frequencies the original is missing anyway.

Where it sits

Stage 4 in the production chain. The declip-repaired source splits into two branches: one goes through a PRE low-pass and into FlashSR (via the external Egregora Audio Super Resolution pack), the other gets cleanly resampled to the FlashSR sample rate. This node recombines them. Its audio output feeds the HF Cymbal / Shimmer Repair stage next.

How it works

It takes two inputs - original_audio and flashsr_audio - and, depending on mode, does one of these:

• Original + FlashSR air (default): the complete clean-resampled original is preserved, and only a linear-phase high-frequency band from FlashSR is added on top. This is the mode you'll actually use.
• Hybrid replace above crossover: original below the crossover, FlashSR above it. More aggressive; lets FlashSR own the whole top end.
• Original SRC only and FlashSR only: pure A/B references, no mixing. Handy when you're trying to figure out what FlashSR is actually contributing.

Three numbers shape the blend: crossover_hz (default 14500 - the center of the linear-phase crossover; roughly 13–15 kHz is a good starting range for a 32 kHz MiniMax source), transition_hz (how soft/wide the FIR crossover is, default 2000), and flashsr_hf_mix (default 0.35 - how much of the FlashSR high band actually makes it through). Lower flashsr_hf_mix values are the safe move when cymbals or reverbs sound smeared; values above 1.0 intentionally exaggerate reconstructed air and are "normally not recommended for mastering," per the tooltip.

Inputs and outputs

The inputs that matter are original_audio, flashsr_audio, mode, and flashsr_hf_mix. Outputs are audio, hybrid_crossover_json (a JSON report of sample rates, crossover parameters, HF mix and mode), and info. Wire the JSON into MiniMax Song Metadata for the sidecar.

Installing it

Part of jplenio's MiniMax Music Production Toolkit - ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/jplenio/ComfyUI-MiniMax-Music-Production-Toolkit.git
cd ComfyUI-MiniMax-Music-Production-Toolkit
python -m pip install -r requirements.txt

The one extra step here: this node does not contain FlashSR. The actual FlashSR stage in the example workflow comes from the separate ComfyUI-Egregora-Audio-Super-Resolution pack, and its model weights are not bundled by this toolkit. If the FlashSR node is missing from your workflow, that's the pack to install (and it runs on GPU or CPU).

Gotchas

If your hybrid result still sounds watery, the README's troubleshooting gives the checklist - and it's worth following in order: lighter PRE filter or bypass → compare Original SRC only → lower flashsr_hf_mix → HF repair on Gentle. Don't assume more reconstructed bandwidth sounds more natural; that's the trap this whole stage exists to guard against.