MixMod Guider Component Pipeline

This is the node that makes the MixMod Pipeline useful. Where the plain MixMod Guider Component takes a raw model and conditionings, this one takes a PIPELINE bundle - the model-plus-prompts you stashed in a MixMod Pipeline node - and turns it into a weighted, CFG'd COMPONENT you can chain into the mix. Same job, neater input.

You'll reach for it the moment you're mixing more than one checkpoint and you've realized you don't want to re-wire the same three connections for every model. Each pipeline becomes one component: pull the pipeline out, set how loud it is, set its CFG, chain it to the previous component via prev_component, done. Swap the checkpoint or edit the prompt inside the Pipeline node and every component built from it updates without you touching the wiring.

The inputs that matter

• pipeline - the PIPELINE from a MixMod Pipeline node. This supplies the model, positive, and negative.
• base_weight (default 1.0, 0–1) - this component's share of the blend. Remember weights get normalized across the whole chain, so these are relative volumes, not absolutes.
• cfg (default 7.5) - guidance scale for this specific model's prediction, applied every step.
• prev_component (optional) - chain the next component in here to stack models.
• options (optional) - an OPTIONS bundle for this component: mask, step range, or scale.

Output is a COMPONENT that feeds the MixMod Guider or a special guider, which converts the chain into a GUIDER for the sampler.

What's happening under the hood

It's a thin adapter: the node unpacks pipeline["model"], pipeline["positive"], and pipeline["negative"] and builds exactly the same {model, positive, negative, weight, cfg, options, prev} dict the plain component node makes. So everything that's true of the MixMod Guider Component - weight normalization, per-model CFG, VRAM doubling from holding each checkpoint loaded - is true here too. There's no functional difference between the two component nodes; this one just reads a bundle instead of loose wires.

Installing

Part of the ComfyUI-MixMod pack. ComfyUI Manager → search "ComfyUI-MixMod" → install and restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/kantsche/ComfyUI-MixMod

No requirements.txt, no model downloads - the pack is plain torch plus ComfyUI's own sampling code. Pair it with ComfyUI-GGUF and ComfyUI_ExtraModels if you're mixing in Pixart Sigma, and the ostris sdxl-sd1-vae-lora if you're mixing SD1.5 with SDXL (that one's in the README and it's the difference between a coherent blend and mush). Author's setup is Windows, ~8GB VRAM floor, so budget your RAM accordingly.

Troubleshooting

If a component you built from a pipeline seems to have no effect, check you didn't set its base_weight so low it normalizes away next to a loud neighbor, and confirm the prev_component chain actually reaches the guider - an orphaned component never gets sampled. If output turns to noise, the classic cause is a latent-space mismatch between mixed checkpoints, which is the vae-lora problem above, not this node.