Inputs Refine Basic

AUNInputsRefineBasic is the "Basic" version of the refine idea: it keeps the lighter Inputs Basic contract - checkpoint, sampler settings, empty latent - and adds one thing, an optional second model for a refinement pass. No save-prep outputs, no legacy filename sockets. It's the recommended-shaped node for anyone who wants the "generate then refine with a second checkpoint" pattern without the full Inputs node's wall of save machinery.

What it does

Load the main checkpoint with ckpt_name, then optionally refine_ckpt - select 'None' to reuse the main model for refinement. When you pick a separate refine checkpoint you get a second MODEL REFINE output (and ckpt name refine) to feed a refinement sampler. The rest is the Basic contract: sampler/scheduler/cfg/steps, aspect_ratio/width/height with aspect_mode and megapixels/multiple, batch_size, seed, clip_skip, and the SpeedLoRA group.

The SpeedLoRA handling is where refine nodes earn their keep: speed_lora_full_both applies the full strength to both main and refine models, while speed_lora_ratio splits it (main model gets that fraction, refiner gets the rest). Outputs include the resolved speed lora ratio, so you can see what landed where. A SpeedLoRA tuned for composition can fight a refiner; splitting it is often the difference between a cleaner final pass and a muddy one.

Inputs and outputs

Inputs: ckpt_name, refine_ckpt, SpeedLoRA group, clip_skip, sampler/scheduler/cfg/steps, aspect_ratio/width/height/aspect_mode, batch_size, seed, megapixels, multiple.

Outputs: MODEL, CLIP, VAE, MODEL REFINE, ckpt name, ckpt name refine, sampler, scheduler, cfg, steps, latent, width, height, seed, batch size, speed lora ratio.

Installing it

Same pack, same install:

• ComfyUI Manager: search "AUN ComfyUI Nodes", install, restart.
• Manual: cd custom_nodes && git clone https://github.com/loz2754/AUN-ComfyUI-Nodes, then restart.

Manual installs and a ModuleNotFoundError: cv2 error? pip install -r custom_nodes/AUN-ComfyUI-Nodes/requirements.txt.

Common issues

If an old workflow using this node loads with SpeedLoRA widgets that look off, the README's migration note covers it - the input set changed and widgets may need re-checking or reconnecting. Otherwise the same rules as the other Basic nodes: this node has no filename outputs (plan save naming separately), and when the refiner misbehaves, adjust the speed_lora_ratio split before chasing sampler settings.