RALUConditioningAdapter

RALUConditioningAdapter is the pack's conditioning shim, and it's honest about how small its job is: it decides what happens to your CONDITIONING before it reaches a RALU sampler. In practice that means one of two things - pass your conditioning through untouched, or stamp it with a guidance value plus a few RALU metadata keys. That's the whole node.

So why would you care? Because the RALU runtime reads guidance out of the conditioning metadata, not from a separate widget. FLUX workflows normally set guidance with a dedicated FluxGuidance node, but SD3 has no such node, and if your conditioning doesn't carry a guidance value the sampler silently falls back to 3.5. If you want SD3's RALU path to use a specific guidance number, this node is how it gets in.

What it actually does

The source is a one-liner. In adapt_conditioning, the mode input picks the behavior:

• passthrough (default) - returns your conditioning exactly as-is. Nothing is added or changed.
• annotate_placeholder - calls ComfyUI's conditioning_set_values and writes four keys onto the conditioning: guidance, plus ralu_placeholder, ralu_conditioning_mode, and ralu_family for the pack's own bookkeeping.

The other inputs are guidance (a float, default 3.5) and family (generic / flux / sdxl / sd3). Family here is only a tag on the metadata - it doesn't change how conditioning is encoded. The single output is a CONDITIONING that plugs straight into RALUSamplerCore, SD3LowResDenoise, or any of the SD3 two-stage nodes.

The honest take

You'll rarely wire this node up yourself. RALUFluxWorkflowBridge already runs adapt_conditioning internally with guidance 3.5 when your conditioning is missing it, so the FLUX path doesn't need it at all. The node is mainly useful when you're hand-building an SD3 chain with SD3LowResDenoise and want a specific guidance value in there. If you're on the bridge, skip it - passthrough is the right answer by default.

Installing

The pack is comfyui-ralu-sampling. Install it once and you get all eight nodes:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/Milkyawaway/comfyui-ralu-sampling.git
cd /path/to/ComfyUI
python -m pip install opencv-python numpy typing-extensions
python main.py

ComfyUI Manager works too - search "comfyui-ralu-sampling" and restart. The package targets ComfyUI's newer node API (comfy_api.latest), so if the nodes don't appear in your palette, your ComfyUI install is too old; update it first.

Troubleshooting

• Set mode to passthrough and the output is identical to the input. If that looks like a no-op, it is. annotate_placeholder only matters when the downstream sampler reads guidance metadata - which means the SD3 nodes and, internally, the FLUX bridge.
• Wrong family tag won't error - it just writes a slightly misleading metadata value. The actual family enforcement happens in the sampler (generic and sdxl throw "not implemented yet"; only flux and sd3 actually run).
• No model files involved. This node only touches conditioning, so it needs nothing beyond the pack itself.

If your SD3 output looks off, the first thing to check isn't this node - it's whether your conditioning carried any guidance at all. And if it didn't, well, now you know the one place to fix it.