GodnessMerger_RAW_Apply

If you've read any of the other Godness Merger node articles, you've hit the same asterisk each time: those nodes generate a ratio, but they don't merge anything themselves - that happens in a combiner node this batch of briefs doesn't cover. GodnessMerger_RAW_Apply is the one node in the family that actually does something with two live models, and it's built around a different idea entirely: instead of generating fresh ratios, it applies a merge recipe you already have in text form.

What it does

Wire in model1, model2, a strength multiplier, and a multiline text field, and it hands back a merged MODEL plus a RATIO_VALUES string. The name is the tell: this is the "RAW" apply - you're not asking it to invent a merge, you're feeding it a raw specification and telling it to execute that exact recipe. The most obvious source for that text is the VALUES string that Random_Model_Merge outputs after a run - that node's whole reason for reporting its chosen ratios back to you is so you can capture a good result and reproduce it later, possibly against a different pair of models, without relying on the same seed landing the same way twice.

Here's the honest gap: nowhere in the SP-Nodes README - which is screenshots, not documentation - does it spell out the exact text format this field expects. I didn't find community discussion of it either; this is a genuinely obscure corner of a niche pack (searching for the pack and the author turned up almost nothing beyond the author's other, unrelated project, a Windows app called Comfizen for building simplified UIs over ComfyUI API workflows - useful for sizing up who built this, not for the text format). The safest way to fill that field is to actually generate a VALUES output from Random_Model_Merge first and paste it in verbatim, rather than trying to hand-write the syntax from scratch.

The inputs that matter

• model1 / model2 - the two checkpoints being merged. Same architecture family required - this is a raw weight-level merge, not something that tolerates mixing SD1.5 with SDXL.
• text - the recipe. Paste a VALUES/RATIO_VALUES string captured from a previous merge, don't hand-type it unless you know the format.
• strength (0–1, default 1) - an overall multiplier on top of whatever the recipe specifies.

Output: MODEL (wire it wherever a checkpoint goes) and RATIO_VALUES (a string echoing back what was actually applied - useful for a paper trail if you're iterating).

Installing SP-Nodes

Manager: search "SP-Nodes," install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/bananasss00/ComfyUI-SP-Nodes

Restart after cloning. No model download, no extra dependency for this node - it operates on models already loaded elsewhere in your graph.

Where this goes wrong

Two failure modes worth knowing about. First: architecture mismatch between model1 and model2 - merging across families (SD1.5 into SDXL, or either into Flux) isn't something this node can gracefully handle; expect an outright error or a broken model, not a warning. Second, and more specific to this node: pasting malformed or hand-typed text that doesn't match whatever format the recipe parser expects. If the merge comes out wrong (or the node errors on the text field), the fastest debugging path isn't guessing at syntax - run Random_Model_Merge once, copy its VALUES output verbatim, and confirm RAW_Apply reproduces the same result from that. Once you know a real example works, you can start editing it deliberately instead of reverse-engineering the format from nothing.