ComfyUI-latent-ops
A collection of nodes for manipulating LATENT in ComfyUI.
Nodes (46)
A tripwire for latent shapes that fails loudly instead of weirdly
Pin the exact latent shape and stop guessing why downstream nodes break
The invisible number that wrecks ELLA prompts
Peek inside a 0-1 latent without paying for a VAE decode
The visualizer for -1..1 latents (it scales, the 01 node doesn't)
Add two latents together — but only when something applies the op
Nudge every latent value by a constant
Re-apply guidance to an already-sampled latent
The ELU variant that stays smooth below zero
Keep latent values inside a range
Clip the ceiling on your latent
Put a floor under your latent
The hyperbolic curve you'll reach for exactly once
The activation that keeps negatives alive
Reset a latent to a constant
The activation your diffusion model already loves
Sigmoid's fast, piecewise cousin
The cheap, smooth-sounding Swish
Tanh with training wheels, and a -1..1 guarantee
The zero-input node that normalizes style away
Blend two latents, the way morphing was always meant to work
When your latents need a recalibration, not a rescale
The ReLU that doesn't murder your negatives
It's log-sigmoid, not the sigmoid you expect
The smooth, self-regularizing activation for latents
Multiply one latent by another latent, element by element
The humble scalar multiply that does most of the work
One global z-score to pull a latent back to earth
Normalize the channel, not the whole world
Squash your latent into 0–1 and hope nothing is an outlier
Min-max normalization that ignores the weird values
The nuclear option for latents, and why you should think twice
ReLU with a cap that your latent will almost never hit
How to break a latent's shape (and maybe your workflow)
Shift your latent around and wrap it back like a scrolling marquee
The self-normalizing activation that mostly normalizes for you
The real S-curve, for squashing latents into 0–1
The modern neural-net activation, applied to your latent
The steepest, most aggressive nonlinearity in the box
Carve a chunk out of your latent — and learn the '0 means end' quirk
Turn a latent axis into probabilities, and wonder why
Softmin is the softmax that cheers for the smallest value
The smooth ReLU that never quite zeroes out
The tanh that takes its time at the edges
Pry the cond and uncond apart when one latent is secretly two
Tanh your latent before it does something dramatic
ComfyUI-latent-ops
A collection of nodes for manipulating LATENT in ComfyUI.
All nodes are added under hnmr/latent_ops.
Nodes
| Name | Function |
| --- | --- |
| AssertDims | Verify that the input LATENT has the specified number of dimensions |
| AssertShape | Verify that the input LATENT has the specified shape |
| LatentOperationNormalizeAlongAxis | Normalize along a specific axis |
| LatentOperationNormalize | Normalize across the entire tensor |
| LatentOperationLayerNorm | Apply layer normalization |
| LatentOperationInstanceNorm | Apply instance normalization |
| LatentOperationNormalizeMinMax | Normalize between minimum and maximum values |
| LatentOperationNormalizePercentile | Normalize based on percentiles |
| LatentOperationSigmoid | Apply sigmoid function |
| LatentOperationHardSigmoid | Apply hard sigmoid function |
| LatentOperationLogistic | Apply logistic function |
| LatentOperationTanh | Apply tanh function |
| LatentOperationHardTanh | Apply hard tanh function |
| LatentOperationSinh | Apply sinh function |
| LatentOperationCosh | Apply cosh function |
| LatentOperationReLU | Apply ReLU function |
| LatentOperationReLU6 | Apply ReLU6 function |
| LatentOperationLeakyReLU | Apply LeakyReLU function |
| LatentOperationELU | Apply ELU function |
| LatentOperationSELU | Apply SELU function |
| LatentOperationCELU | Apply CELU function |
| LatentOperationGELU | Apply GELU function |
| LatentOperationSiLU | Apply SiLU function |
| LatentOperationHardSwish | Apply HardSwish function |
| LatentOperationMish | Apply Mish function |
| LatentOperationSoftplus | Apply Softplus function |
| LatentOperationSoftmax | Apply Softmax function |
| LatentOperationSoftmin | Apply Softmin function |
| LatentOperationSoftsign | Apply Softsign function |
| LatentOperationReshape | Change the shape of LATENT |
| LatentOperationSlice | Extract a portion of LATENT |
| LatentOperationRoll | Shift LATENT |
| LatentOperationAddBroadcast | Add values with broadcasting |
| LatentOperationMulBroadcast | Multiply values with broadcasting |
| LatentOperationFill | Fill with specified value |
| LatentOperationAdd | Add another LATENT |
| LatentOperationMul | Multiply by another LATENT |
| LatentOperationClamp | Limit values with upper and lower bounds |
| LatentOperationClampMin | Limit values with lower bound |
| LatentOperationClampMax | Limit values with upper bound |
| LatentOperationApplyCFG | Apply CFG adjustment |
| LatentOperationSplitCFG | Split CFG into conditional and unconditional parts |
| LatentOperationInterpolate | Interpolate between two LATENT tensors |
| Latent01ToImage | Convert LATENT in 0-1 range to image |
| Latent11ToImage | Convert LATENT in -1 to 1 range to image (z = x * 0.5 + 0.5) |
| GetSigma | Get σ at a specific index from SIGMAS |