Blender Output Save Image

This is the node every image-generating workflow ends at. BlenderOutputSaveImage is ComfyUI's standard SaveImage node, subclassed so it slots into the Blender bridge - it saves the incoming IMAGE to the server's output directory and, crucially, reports it back to the add-on so the result can be pulled into Blender. Without an output node in the graph, the Blender add-on has nothing to grab; this is how your generated image becomes a texture, a background plate, or a reference in the viewport.

Mechanically it's thin: INPUT_TYPES calls the parent's, then overrides the default filename_prefix to blender. The saving itself is 100% native ComfyUI. The Blender-side magic is that the add-on watches for this node's output in the run results and offers the saved file for loading into Blender - which is why "run the workflow in ComfyUI once to confirm it executes" is step two of the pack's setup, before you ever import into Blender.

Inputs and outputs

• images - the IMAGE tensor to save. Wire this to your VAE Decode output (or wherever pixels come from).
• filename_prefix - the folder/filename prefix for the saved file, default blender. Set it to something like textures/crate and files land in output/textures/crate_00001_.png. Organizing prefixes by use case is the single easiest way to keep results findable.

It's an output node, so there are no meaningful outputs - though the schema does pass the IMAGE through on an images output socket, so you can chain it into a preview or a second saver if you want.

Installing it

Standard for the pack. ComfyUI side, ComfyUI Manager → search "ComfyUI-Blender", or:

cd ComfyUI/custom_nodes
git clone https://github.com/alexisrolland/ComfyUI-Blender

Restart ComfyUI. No extra Python dependencies, but you need a current ComfyUI - schema v3 nodes, latest required per the README. The Blender add-on is the other half, a ZIP from the latest release installed via Edit > Preferences > Add-ons > Install from Disk.

Where people get burned

Files save to the server's output folder. With a remote ComfyUI, that's not your Blender machine's disk - the add-on fetches the result over the network, so make sure the server is reachable and (if it's on another machine) started with --listen. Second, the prefix defaults to blender, and if every output node in every workflow shares that prefix you get an unreadable pile of blender_00001_.png files; prefix by project. Third, the classic full-pack flow mistake: exporting a normal workflow JSON instead of API format, which means the add-on never sees the node at all. And if Blender and ComfyUI share one GPU, remember the viewport's VRAM is part of the same budget your VAE decode is asking for.