EditUtils: Boogu Output Extractor lrzjason

The Boogu encode nodes in this pack hand you a custom_output socket typed as ANY. That socket is a dictionary with your workflow's internals inside it - negative conditioning, reference latents, vision-tower images, VAE-encoded images. Useful, sure, but a dictionary is not a wire. BooguOutputExtractor_EditUtils is the tool that unpacks that dictionary into four typed outputs you can actually connect things to.

One required input: custom_output. Four outputs: negative_cond (CONDITIONING), ref_latents (LIST), vl_images (LIST), and main_image (IMAGE). The mechanism is honest and easy to trace in the source - it pulls each key out of the dict and hands it over, with one bit of translation worth knowing: main_image isn't stored as its own key, it's computed as the first entry of the internal vae_images list. So if your Boogu workflow encoded no images (a text-only run, say), main_image comes back as None and anything downstream of it will complain. That's not a bug, that's "you asked for an image that doesn't exist."

Where each output goes in a real workflow: negative_cond is the conditioning the node built for the negative side - feed it to your sampler's negative input and you're honoring Boogu's "refs on both sides" design without assembling anything yourself. ref_latents and vl_images are the lists you dig into when you want to inspect or reuse individual references (pair them with ListExtractor_EditUtils or the Any2Image/Any2Latent converters in this pack to pull one item out and make it a typed wire). main_image is your preview-able starting point - wire it to a PreviewImage or SaveImage to actually see the canvas the sampler will start from.

Honest assessment: for a simple Boogu edit you may not need this node at all - BooguEditTextEncode_EditUtils already gives you positive and negative as first-class outputs, and the extractor is only worth it when you want the guts. It earns its place the moment you do. Want the negative conditioning as a standalone wire to compare or reuse? Want to save the reference latents for a second run without re-encoding? Want to preview exactly what the vision tower is looking at? That's when you insert the extractor between the encode node and the rest of the graph.

One tip from the changelog: a recent update fixed the no_refs_cond output family across the Output Extractors, so if you're on an older pull of lrzjason/ComfyUI-EditUtils and something here doesn't match, update before debugging your graph. Install via ComfyUI Manager (search "EditUtils") or git clone https://github.com/lrzjason/ComfyUI-EditUtils into custom_nodes, restart. No models, no dependencies.