Hard Clamp ConDelta between -strength and strength

The pack ships two clamping nodes, and this is the blunt one. ClampConDelta smoothly squashes values into a fixed -1 to 1 range with nothing to configure; HardClampConDelta gives you a literal min/max cutoff at a bound you choose. Anything past that bound gets truncated flat, no smoothing.

Why you'd reach for it

Same underlying need as ClampConDelta - you've built or loaded a delta with an unpredictable magnitude and you want to cap how hard it can push a generation - but with two things that node doesn't offer: a configurable bound instead of the fixed -1/1 range, and a genuinely hard cutoff rather than a smooth squash. Use this one when you know the specific ceiling you want (say, matching a bound you've already tuned elsewhere in your workflow) and you're not worried about the discontinuity a hard clip introduces at the edge. Use ClampConDelta when you'd rather avoid that sudden edge and don't need a custom bound.

How it works

The display name says it directly: "Hard Clamp ConDelta between -strength and strength." Every value in the conditioning that falls outside [-strength, strength] gets clipped to the nearest bound. Values already inside the range pass through untouched.

The inputs and outputs that matter

• conditioning - the conditioning or ConDelta you're clamping.
• strength - default 1, step 0.01, no explicit min/max in the schema. This sets both bounds symmetrically: the clamp range is [-strength, strength].

Output is a single CONDITIONING.

How to install it

Via ComfyUI Manager: search "ComfyUI-ConDelta", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/envy-ai/ComfyUI-ConDelta

then restart. No model downloads for this node.

Common issues & troubleshooting

A hard clamp can introduce visible artifacts of its own if the bound is too tight. Truncating values abruptly rather than smoothing them means a delta that gets clipped hard can end up looking noticeably different from an unclipped version - not just "less extreme" but structurally altered, because you're flattening a whole tail of the distribution to a single value. If that's showing up in your output, either raise strength or switch to ClampConDelta's smoother squash.

This is a safety rail, not a creative control. Like the rest of the pack's clamping and scaling tools, don't expect a visible effect if your input was already well inside the bound - it only does something once values actually exceed strength.

Pair it with ApplyConDeltaAutoScale's own optional clamp if you're already using that node - the auto-scale nodes have their own built-in clamp with a clamp_value, so you may not need this separate node in that particular chain. Reach for HardClampConDelta specifically when you're building a delta manually, outside the auto-scale nodes, and want the same kind of ceiling.