Prompt Subject Slider

Prompt Subject Slider is the sequel to Prompt Subject, and the pitch is right in the name: every one of its 21 categories gets a weight slider next to it, so instead of just choosing 1girl you can decide how strongly you want 1girl. If you've ever wanted to dial a tag up or down without hand-editing (tag:1.3) syntax, this is the node.

How it works

The mechanics are identical to Prompt Subject - same subject.json data file, same 43-character / 112-subject-type / 272-action dropdowns, same Random handling, same seed input for reproducible dice. The difference is what happens when it builds the string. For each category you set, the node checks that category's <key>weight slider (default 1.0, range 0.1–2.0, stepped at 0.1). If the weight is exactly 1.0, the tag goes in bare: 1girl. If you've pushed it anywhere else, the tag comes out wrapped in the standard ComfyUI weight syntax:

• weight 1.3 → (1girl:1.3)
• weight 0.7 → (1girl:0.7)

So the node is literally writing (tag:weight) for you, which is exactly what a beginner should be learning to do by hand - except the slider makes the fiddly part visual. It applies the same weighting to every category you touch, including Character, Action, Action+, clothing, and the rest, and it applies to Random picks too.

The output is the same single prompt string, comma-joined, wired into a CLIP Text Encode.

The honest caveat about weights

Weights work great on CLIP-era checkpoints - SDXL, Illustrious, Pony, NoobAI, WAI - where (tag:1.3) genuinely shifts attention. The practical sweet spot is roughly 0.5–1.5; below 0.5 the concept barely registers and above 1.5 you slide into oversaturation and artifacts. But on LLM-encoded models (Flux, Chroma, Anima, Qwen-based ones) those weights are basically dead weight: the text encoder treats (tag:1.3) as literal punctuation and it lands in the prompt as noise. So on a modern Flux workflow, leaving every slider at 1.0 (or using the plain Prompt Subject) is the right call, and dragging them up will just confuse the model.

Installing it

Same pack, same drill - ComfyUI-RvTools_v2 via Manager (search "ComfyUI-RvTools_v2") or:

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI-RvTools_v2
# restart ComfyUI

No model files to download; the tag data is in the pack's json folder.

Gotchas

Beyond the weighting-on-LLM-encoders trap above, the main thing to know is that this node produces a lot of dropdowns - 21 categories plus 21 sliders - so it takes up serious canvas space. Group it or fold it into a RvTools group if you want a clean graph. And the pack-level caveat applies: RvTools_v2 is unmaintained, superseded by ComfyUI_Eclipse, and the [RvTools] suffix on node names can trip up ComfyUI Manager when it tries to match nodes from shared workflows.