IPAdapter Mad Assistant V5
The -5 to 5 edition, and the end of the line
- layer_weights
IPAdapter Mad Assistant V5 is the top of the family tree: the twelve layer Weight sliders span -5.0 to 5.0, the widest range in the pack. Every V-number in this series is the same node with a bigger range bolted on, and V5 is where the author stopped expanding.
What's it actually for? A couple of things, honestly. The one defensible use is hammering one or two late style layers hard (think 3–5) while keeping the early structure layers near zero - the reference's look comes through while its composition stays out of the way. The other, less defensible use is random mode, which on this node fills currently-zero layers from the full -5.0 to 5.0 range. That's not "a little variety," that's each zero layer rolling a die between "ignore the reference" and "reference at full blast," with negatives that actively push against it. If you run Random on V5, pin the layers you care about to non-zero values and brace.
Mechanics are family-standard: twelve sliders in, comma-joined layer:weight string out through the layer_weights STRING output, wired into the IPAdapter Mad Scientist node's layer_weights input (right-click → Convert to Input first). And yes, the seed input is a lie - it's declared in the schema but never read by the code, so nothing about Random mode is reproducible.
The honest take on the V-numbers
The version numbering is the pack's least clear feature. V1 is -1..1, V2 is -2..2, V3 -3..3, V4 -4..4, V5 -5..5 - each is just a wider slider. The README adds to the confusion by calling V2 the "-1 to 1" one (it isn't) and describing "0–10 layers" (there are 0–11). So when you see "Mad Assistant V5" in a workflow, read it as "the extreme-range one," not "the newest one" - there's a V6, and it's the same range.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/11dogzi/ComfUI-EGAdapterMadAssistant
Restart ComfyUI, or search ComfUI-EGAdapterMadAssistant in ComfyUI Manager. No pip, no dependencies, no model downloads - a single pure-Python file. You still need ComfyUI_IPAdapter_plus and its models installed for Mad Scientist to exist.
Reach for V5 when the extreme is genuinely what you're testing. For anything resembling normal style transfer, the base node (0..1) or V1 (-1..1) is where you should be living.
Inputs (14)
| Name | Type | Default | Description |
|---|---|---|---|
| 0layer Weight | FLOAT | 0.0-5–5 | — |
| 1layer Weight | FLOAT | 0.0-5–5 | — |
| 2layer Weight | FLOAT | 0.0-5–5 | — |
| 3layer Weight | FLOAT | 0.0-5–5 | — |
| 4layer Weight | FLOAT | 0.0-5–5 | — |
| 5layer Weight | FLOAT | 0.0-5–5 | — |
| 6layer Weight | FLOAT | 0.0-5–5 | — |
| 7layer Weight | FLOAT | 0.0-5–5 | — |
| 8layer Weight | FLOAT | 0.0-5–5 | — |
| 9layer Weight | FLOAT | 0.0-5–5 | — |
| 10layer Weight | FLOAT | 0.0-5–5 | — |
| 11layer Weight | FLOAT | 0.0-5–5 | — |
| Random | COMBO | No | 2 options: Yes, No |
| seed | INT | 0-1125899906842624–1125899906842624 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| layer_weights | STRING | — |