SAM Embedding to Image

This is the small companion to Get SAM Embedding. A SAM_EMBEDDING is an abstract blob of features - Segment Anything's encoder output - and there's nothing to look at in the graph. This node turns that blob back into a viewable IMAGE so you can actually see what got encoded and confirm the segmentation half of your workflow is doing what you think.

It's a debug/preview tool, plain and simple. You won't ship it in a final render, but it's the node that answers "wait, is the SAM step even working?" without you having to run the whole masking pipeline to find out.

How it works

You feed it the SAM_EMBEDDING produced upstream, and it decodes that representation into an image you can drop a Preview Image on. Think of it as a window into an otherwise opaque intermediate - the same way you'd wire a VAE Decode just to eyeball a latent. There's exactly one input and one output, so there's not much to get wrong.

The inputs and outputs

• embedding (SAM_EMBEDDING) - the output of Get SAM Embedding. That's the only input.
• IMAGE (output) - a visual rendering of the embedding, ready for a Preview Image or Save Image node.

That's the whole node. It exists to close the loop on the SAM encode step.

Installing it

Grab the pack through ComfyUI Manager (search comfyui-art-venture, install, restart) or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/sipherxyz/comfyui-art-venture

and restart ComfyUI. On comfy.icu it's already in the shared image.

Common issues

If it loads red as a "missing" node, that's the art-venture pack not being installed - a common situation when you open a shared workflow, since this pack is a frequent behind-the-scenes dependency. Manager → "Install Missing Custom Nodes" fixes it.

Beyond that, there's almost nothing to break. The one requirement is that the embedding input comes from a real SAM_EMBEDDING - you can't feed it anything else, and ComfyUI's type system will refuse the wire if you try. If the preview looks like noise or nothing meaningful, the problem isn't here; it's upstream at the encode step (wrong SAM model, or an image that didn't encode cleanly). Fix it at Get SAM Embedding, not here. This node only ever reflects what it was handed.