preset_selector_custom_node

Preset Selector 10 — ComfyUI custom node

A ComfyUI custom node that stores 10 presets and applies the selected one by index. Each preset bundles a HIGH LoRA, a LOW LoRA, and positive / negative prompts, so you can switch between full setups with a single integer.

Nodes

• Preset Selector 10 — pick 1 of 10 presets and output the LoRA-applied models and encoded conditioning. Each slot's (high_lora, low_lora) pair can auto-load its prompt from a persistent library and be saved back with one button — see Library persistence below.
• Modulo 10 — small helper that wraps an incrementing INT into the 0–9 range.

Files

• __init__.py
• preset_selector.py

Install

1. Copy the folder preset_selector_custom_node into your ComfyUI custom_nodes folder:

   custom_nodes/preset_selector_custom_node/
   

2. Restart ComfyUI.
3. Search for these nodes:
   • Preset Selector 10
   • Modulo 10

Usage

• Connect the base MODEL and CLIP into Preset Selector 10.
• (Optional) Connect a second base model to model_low. When connected, the LOW LoRA is applied to model_low (and the HIGH LoRA to model); when left unconnected, both LoRAs apply to model as before. This lets one node drive two different base models — see Dual base models below.
• Set preset_index.
• Fill the preset_0 … preset_9 fields.

Outputs

| Output | Description | |--------|-------------| | high_model | Base model with the preset's HIGH LoRA applied | | low_model | Base model with the preset's LOW LoRA applied | | positive | Encoded positive conditioning | | negative | Encoded negative conditioning | | selected_index | The resolved preset index (0–9) | | selected_name | The selected preset's name |

preset_index is wrapped with modulo 10, so 10 → 0, 11 → 1, and so on.

Auto-changing the preset per queue run

Option A — manual

• Add an INT (or primitive int) node and set Control After Generate = increment.
• Connect that INT to preset_index, or route it through Modulo 10 first.
• Each queue run advances to the next preset.

Option B — with a batch image loader

• Keep your batch loader's image_index incrementing.
• Keep a separate INT node incrementing for presets, starting both from 0.
• The selector wraps preset_index % 10, so 0–9 cycle cleanly even as the INT keeps rising.

Notes

• The HIGH / LOW LoRA fields are dropdowns populated from your ComfyUI loras folder (e.g. MimicPC's models/loras). Pick None to skip the LoRA for that slot.
• The dropdown list is read when ComfyUI loads. After adding files to models/loras, restart ComfyUI (or reload the node) to see them.

Library persistence

Each of the 10 slots can auto-load its prompt from — and save it to — a persistent per-user library keyed by the slot's (high_lora, low_lora) pair, so you don't re-type prompts across workflows.

• Changing a slot's high_lora / low_lora auto-fills that slot's positive / negative / strengths / name from the library (if that pair was saved). No saved entry → the slot is left unchanged. Opening a saved workflow never overwrites its slots.
• 💾 Save all slots to library upserts every slot that has a LoRA pair (skipping empty None/None slots) back into the library in one write.
• Loaded values are editable in place; execution uses the on-screen values (what you see is what runs). The library is stored via ComfyUI's userdata API as preset_selector_library.json in ComfyUI's user directory, so it survives node updates and persists on MimicPC.

Using on MimicPC

MimicPC runs stock ComfyUI (plus ComfyUI Manager) in the browser, so install and usage are the same as any ComfyUI. Two things are MimicPC-specific: LoRAs are read from Storage > models > loras (persistent), and this node's library JSON lives in ComfyUI's user/ directory — the same persistent area as saved workflows — so your presets survive instance restarts and node updates.

Prerequisites

• A ComfyUI app running on MimicPC.
• Your LoRA files uploaded to Storage > models > loras (the dropdowns read from there).

Install

Option A — ComfyUI Manager (recommended):

1. In ComfyUI, open Manager → Install via Git URL.
2. Paste https://github.com/chikiu-san/preset_selector_custom_node and click Install.
3. Restart the ComfyUI app.

Option B — manual: place this folder under ComfyUI's custom_nodes/ (clone or upload), then restart.

Use it

1. Right-click the canvas → Add Node → presets → Preset Selector 10 (or double-click and search).
2. Connect MODEL and CLIP (from your checkpoint loader).
3. In any slot, pick high_lora + low_lora. For a new pair, type the prompts (and a name), then click 💾 Save all slots to library.
4. Next time — in any workflow — set a slot to the same pair and its saved prompt / strengths / name auto-fill. Tweak in place; press 💾 again to update. Use preset_index (with Control After Generate = increment) to cycle through the 10 slots as before.

Notes & troubleshooting

• The LoRA dropdowns are read at ComfyUI startup — after adding files to models/loras, restart to see them.
• The library persists at user/preset_selector_library.json on MimicPC (survives restarts and node updates).
• Node missing after install → confirm you restarted ComfyUI, and check the Manager log for errors.
• Auto-fill does nothing → you haven't 💾-saved that exact (high_lora, low_lora) pair yet (an unsaved pair leaving the widgets untouched is expected behavior).

Dual base models (Wan 2.2)

Wan 2.2 denoises in two stages that use two different base UNet models (high-noise and low-noise), each needing its own LoRA. Connect the optional model_low input so a single Preset Selector 10 drives both:

high-noise base ─→ model       ┐
low-noise base  ─→ model_low   ┤ Preset Selector 10 → high_model → high-noise sampler
                               │                     → low_model  → low-noise sampler
                               └                     → positive / negative → your conditioning

• high_model = model + high_lora
• low_model = model_low + low_lora (falls back to model if model_low is unconnected)
• One prompt, one preset slot, preset_index cycling — all as usual.

Each preset slot bundles a character's _HIGH and _LOW LoRA files as one pair, so selecting a slot applies the right LoRA to each base model.

License

MIT