Donut Filler Model (DEPRECATED)

DonutFillerModel outputs a single model - and it is not a model. It's a placeholder object carrying a marker that this pack's merge code recognizes as "skip this slot." Take nothing in, put a fake model out, and let the graph stay fully wired when a merge node has more inputs than you have models to fill them.

This is the model-side half of the pack's stub family. The combined DonutFiller node emits both model and clip, DonutFillerClip covers just the CLIP, and this one covers just the model. All three do the same thing - emit an object marked _is_filler so the WidenMerge UNet machinery skips it instead of trying to merge a null into your weights.

The realistic use case is narrow and specific to this pack's merging workflow: you're building a merge graph with several optional model slots (model_3, model_4, ...), and you want the graph to stay connected whether you're merging two models or six. ComfyUI's unconnected optionals already arrive as None, and the merge code filters those - so a filler is mainly for template workflows where something forces a connection, or where a preset swaps model counts and you want every socket pre-populated with a harmless no-op.

The honest verdict: this node is DEPRECATED (the combined DonutFiller is what the pack wants you to use), it's a stub, and you should not use it to generate images. If you feed its model output into a sampler you'll get nothing useful - the marker means nothing outside the pack's merge code, and whatever sampler you wire it into will just choke on an empty shell. If a downloaded workflow contains it, you can almost always delete it and reconnect the real model wire.

Install: ships with the pack - ComfyUI Manager → search DonutNodes, or git clone the repo into custom_nodes/ and pip install -r requirements.txt with the same Python that runs ComfyUI. No model downloads, no settings. It's a constant; there's nothing to tune, nothing to re-run, and nothing to get wrong except expecting it to do something.