SettingParameters01

Where LoadPresetForSetting01 just picks a preset file, SettingParameters01 is where the values actually land. It shows the whole preset as a row of widgets and fires them out as outputs you wire into your existing loader and sampler stack. If you use this pack as a pair, this is the node doing the real work.

It's a display-and-forward node, not a mega-loader. The checkpoint stays in your normal CheckpointLoaderSimple, prompts get encoded by your usual CLIPTextEncode nodes. This node just fills in the blanks from a preset, so you can stop re-entering the same sampler settings on every model swap.

How it works - and where people get burned

The node has an optional preset_data input (TOMLDATA) fed by LoadPresetForSetting01. When it's connected, the node reads the preset at queue execution time and returns the preset's values. It also sends a websocket message to the frontend that rewrites what the widgets display.

Here's the trap: because the preset is re-applied every time you queue, any manual tweak you make to a widget gets stomped on. Change ClipNum to -1 and the run still uses the preset's -2. The README's advice is to disconnect or bypass the preset node to keep an edit - easy to forget, which is exactly why the author also made an integrated version. Leave preset_data unconnected and the node just echoes whatever's in its widgets, which is a handy manual fallback.

Outputs that matter

• POS A / POS C / NEG A / NEG C - four text prompts. The Pony sample splits the score tags (score_9, score_8_up…) from the style tags, so wire A and C into separate CLIPTextEncode nodes.
• ckpt_name → CheckpointLoaderSimple's ckpt_name input.
• stop_at_clip_layer (INT) → CLIPSetLastLayer. It's negative-numbered, default -1; -2 is the classic SD 1.5 anime/Pony value, and on SDXL the encoder already skips a layer, so don't over-tune it.
• vae - sneaky one: despite the stringy name it's a fully-loaded VAE object (the node reads the file and constructs it), so it plugs straight into VAEDecode.
• Steps / CFG / sampler_name / scheduler → KSampler.

Getting those into KSampler needs the standard ComfyUI trick: right-click KSampler → Convert Widget to Input → convert the steps, cfg, sampler_name and scheduler widgets. Same for CheckpointLoaderSimple's ckpt_name.

Install

Via ComfyUI Manager (search "ComfyUI_mittimiLoadPreset"), or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/mittimi/ComfyUI_mittimiLoadPreset

Restart ComfyUI. Only dependency is toml, no models to download.

Troubleshooting

"I changed the widget but the run ignored it" is the queue-time overwrite above, not a bug - bypass the LoadPreset node. And before you invest: the README's banner says this pack is superseded by ComfyUI_mittimiLoadPreset2 and won't be updated. The split design works fine for existing workflows, but new ones might as well start on v2.