Preset Settings Loader

Some checkpoints want 512x768, CFG 8, dpmpp_sde karras, 30 steps. Others want 832x1216, CFG 4, euler, 20 steps. If you bounce between models, retyping that wall of settings into a fresh workflow every time gets old fast - that's the exact pain the A1111/Forge "Config Presets" extension solved, and Preset Settings Loader brings it to ComfyUI.

It's a single node with one dropdown. Pick a preset name and it spews out twelve outputs: width, height, CFG, steps, sampler, scheduler, and a block of hires-fix values. All from a JSON file - settings as data, not settings as twelve widgets you have to hunt down across the graph.

How it works

The loader reads configs/generation_presets.json inside the pack folder. On the first ever run that file doesn't exist, so the node creates one with a default "dreamshaper 8" preset (512x768, CFG 8, dpmpp_sde / karras, 30 steps). From then on, whatever you - or the pack's Saver node - write into that file becomes a dropdown entry. It's a plain lookup unpacked into outputs: no execution, no side effects, just values.

The outputs that matter

The four you'll actually use first:

• width, height - into an EmptyLatentImage node.
• cfg, steps - into the KSampler.
• sampler_name, scheduler - also into the KSampler; the sockets are typed *, so ComfyUI lets them connect to any matching input.

Then there's the hires cluster: hires_steps, hires_cfg, hires_denoise, hires_upscale_by, hires_model, and is_latent. Those only matter if you're building a two-pass hires-fix graph - for a first workflow, ignore them and take the first six. The whole pitch is "change one dropdown instead of remembering ten numbers."

Installing

Same pack as everything here. ComfyUI Manager, search "ComfyUI Config Presets", or:

cd ComfyUI/custom_nodes
git clone https://github.com/pooria98/ComfyUI-Config-Presets.git

then restart ComfyUI. No model files, and no heavy dependencies - the pandas/openpyxl in this pack's requirements are only needed by the style-selector node, not this one.

Common issues

• Presets don't show up. If you just saved one via the Saver node, you must refresh the page - this loader builds its dropdown at load time and won't notice a file change until then. The README says this explicitly, and it trips people every single time.
• Outputs look generic. If a hand-edited preset is missing a field, the node silently falls back to defaults (512x512, CFG 7, euler / normal) rather than erroring. If width/height come out as 512 when they shouldn't, open the JSON and check for typos.
• It doesn't load your checkpoint. The preset is numbers and names only - a "dreamshaper 8" preset won't load the checkpoint itself. Keep your model loader wired up; this is the settings half of a saved config, not the whole thing.

The take

If you mostly run one checkpoint at one resolution, this node solves a problem you don't have. But if you juggle several models, or you want to share "here's exactly what produced this image" as a named preset instead of a wall of screenshots, it's quietly the most useful node in the pack.