InjectNoiseToLatent (🚫Deprecated)
Deprecated, and the README tells you what to use instead
- pipe_to_noise
- image_to_latent
- latent
- noise
- mask
- LATENT
The node's own category in ComfyUI is literally EasyUse/🚫 Deprecated - the pack author isn't hiding the ball here. If you're landing on this page because it showed up in a search or an old workflow, the short version is: don't build anything new around it. The pack's own changelog is explicit that noise injection now lives on easy preSamplingNoiseIn (and easy latentNoisy) instead, folded into the normal presampling flow rather than sitting as a standalone latent-mangling node.
What it did, for context
If you've got an old workflow using it, here's what's happening: it takes a latent and adds scaled random noise directly to it, outside of any sampling step. strength (default 0.1, but the range goes absurdly high to 200) controls how much noise gets mixed in - at low values it's a gentle nudge, at high values it's closer to replacing the latent with noise entirely. normalize and average are post-processing toggles on how that mixed noise gets scaled back down. mix_randn_amount blends in a separate batch of fresh random noise on top of whatever you fed in, and seed makes that reproducible.
The optional inputs show it was built to be flexible about where the latent comes from: pipe_to_noise pulls a latent straight out of an Easy-Use pipe, image_to_latent lets you feed a raw image and have it VAE-encoded first, latent and noise take latents directly, and mask restricts the injection to a masked region only. Output is a plain LATENT.
Why it got deprecated
Standalone noise injection nodes like this one have an awkward property: they have to sit in front of your sampler as a separate step, which means every workflow using one needs an extra box wired in exactly the right place, easy to misplace or forget to update when you restructure a graph. Folding the same capability into the presampling config node (which you already have in any Easy-Use workflow) removes that extra wiring entirely - you just set a factor on the node you're already using. That's a strictly better shape for the same feature, which is exactly why this one got flagged for retirement rather than removed outright.
Installing it
If you specifically need it for an old workflow: ComfyUI Manager, search ComfyUI Easy Use, install, restart. By hand:
cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use
Windows: double-click install.bat. Linux/Mac: run pip install -r requirements.txt yourself after cloning. Restart ComfyUI. It ships with the rest of the pack - there's no separate install path, deprecated or not.
What to actually do
Swap it for easy preSamplingNoiseIn if you're already using Easy-Use's presampling pattern (you almost certainly are, if this node showed up in your graph). Its factor input does the same job as this node's strength, just scaled to a saner 0–1 range and applied as part of your normal sampling configuration instead of as a separate pre-sampler mangling step. If you're loading an old shared workflow that still has this node in it, the safest move is to open it, note the strength and seed values, delete the node, and rebuild that same noise nudge on the presampling node instead - don't assume it'll keep working forever just because it still loads today.
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| strength | FLOAT | 0.10000–200 | — |
| normalize | BOOLEAN | false | — |
| average | BOOLEAN | false | — |
| pipe_to_noiseopt | PIPE_LINE | — | |
| image_to_latentopt | IMAGE | — | |
| latentopt | LATENT | — | |
| noiseopt | LATENT | — | |
| maskopt | MASK | — | |
| mix_randn_amountopt | FLOAT | 0.0000–1000 | — |
| seedopt | INT | 1230–18446744073709550000 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |