Latent Image Parameter Bus

img2img has a standard shape. You encode a source image to a latent, then run it through a KSampler that needs a seed and a denoise value. Three things - the latent, the seed, the denoise - coming from three different places, all feeding one sampler. The Latent Image Parameter Bus collects them into a single node and hands each back out as its own typed output. The README's description says it best: "a latent, seed, and denoise for an image2image KSampler source."

It's a pure pass-through, true to the pack's design. The seed and denoise inputs are widgets by default, so the node doubles as a tidy control panel: set your seed and denoise here, wire the latent in from your VAEEncode, and the whole "source" of your img2img branch is one easy-to-read node instead of three scattered connections.

The inputs and outputs

• latent (LATENT) - required. Usually the output of VAEEncode fed from your source image.
• seed (INT) - default 0, full 64-bit range. Goes to the KSampler.
• denoise (FLOAT) - default 1.0, range 0–1, step 0.01. At 1.0 the image is essentially redrawn; lower values keep more of the original structure, which is how people do light "remix" passes.

All three come back out under the same names - latent, seed, denoise - and all three plug into the KSampler's matching inputs. For the scale side of the same workflow, the companion node is the Image Scale to Side Parameter Bus, which holds the side length, seed, and denoise of the scale-to-side variant. If you need both, they make a tidy pair at the head of an img2img branch.

Installing it

Part of the comfyui-bus-plugin pack with the other ten bus nodes. ComfyUI Manager: search "comfyui-bus-plugin", Install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/rhdunn/comfyui-bus-plugin

then restart. Zero dependencies, no model downloads, nothing to configure.

Common issues

The recurring mistake is expecting the denoise to be applied here. It isn't - this node only holds the number; the KSampler does the work. If your img2img looks unchanged no matter what denoise you set, check that the sampler actually receives this value (and that the sampler's own denoise field, if you left one in, isn't overriding it). Also note the seed default of 0: run twice with the same seed and same latent and you'll get the same image, which surprises people who expected variation. That's the seed doing its job, not a bug.