LoadSetParametersLite

If you've ever re-dialed the same checkpoint, prompts, size, sampler, and steps into a workflow you've shared, this is the node that makes you stop. LoadSetParametersLite (that's its menu name in the mittimiTools category) is the hub of the mittimiLoadPresetLite pack: one node with every generation setting on it, plus a dropdown that swaps all of them at once by loading a preset file. Pick "sample.toml" and the whole graph re-dials to the SDXL values that preset ships with - 832×1152, euler_ancestral, CFG 6.5, 32 steps. Pick another preset and it re-dials again.

The 19 widgets read like a checklist of everything a txt2img run touches: checkpoint, clip_num, VAE, three positive prompt fields (A/B/C), three negative fields, width, height, batch size, steps, CFG, sampler, scheduler, seed - and the preset dropdown at the end. The A/B/C prompt split is a plain text concatenation (positive_A + positive_B + positive_C), so you can keep a stable style prefix in A, a subject in B, and a quality suffix in C, then swap only the middle slot between presets.

How the preset swap works is a neat little round trip. When you change the dropdown, the pack's JavaScript posts the preset name to a backend route, which parses the matching .toml file from the pack's presets folder and broadcasts the values back into the node's widgets. All switching happens in the UI - the node itself just reads its widgets and passes them out. Two things follow. First, the preset list is built once when ComfyUI starts, so a preset you save won't appear in the dropdown until you restart. Second, presets don't store the seed - the TOML format has no Seed field - so loading a preset snaps the seed back to 1. Keep an eye on that when you're chasing a specific result.

What comes out: 14 outputs, and only a handful get wired. checkpoint_name feeds LoadCheckpointLiteMittimi; positive_prompt_text and negative_prompt_text feed the encoding path; steps, cfg, sampler_name, scheduler, and seed feed the KSampler; width, height, and batch_size feed EmptyLatentImage; and parameters_data (the pack's PDATA object) feeds SaveImageWithParamTextLite so the saved file records the whole recipe.

Here's the "Lite" part, and it trips people up. This node does not load the model. In the full ComfyUI_mittimiLoadPreset2 pack (which the README points you to for usage docs), the equivalent node loads the checkpoint internally; in this pack that loading code is commented out and split into a separate LoadCheckpointLite node. You wire the name/clip/vae outputs here into that node, and it does the heavy lifting. Cleaner separation once you're used to it - the param source is pure data, the loader is pure compute - but if you expected one node to do everything, that's why nothing renders.

Install through ComfyUI Manager (search "ComfyUI_mittimiLoadPresetLite") or:

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

then restart ComfyUI. The only dependency is toml, and no model files are downloaded. One known conflict: if you run RES4LYF, the README says the pack "will not function properly" and includes a three-line patch to add to RES4LYF's init.py (registering bong_tangent and beta57) so the two packs stop fighting over the scheduler list. If your scheduler dropdown looks wrong after installing both, that's your first suspect.

Final take: if you live in one workflow and never swap settings, skip this - stock nodes do the same job with less ceremony. If you keep a dozen "recipes" around and hate re-entering them, this is exactly the node to reach for.