Inputs Diffusers

Modern diffusion models increasingly ship as split files - a standalone UNet (or transformer), a separate CLIP text encoder, and a separate VAE - instead of one combined checkpoint. Flux is the famous example. AUNInputs is built for combined checkpoints; AUNInputsDiffusers is its cousin for the split world: it loads a diffusion-only UNet plus explicit CLIP and VAE files, and then emits the same downstream inputs as the full AUNInputs node, including the save-prep outputs.

So the pitch is identical to AUNInputs - one node replacing the loader stack plus the sampler settings plus the latent creation - but the model loading is the diffusion-style contract: diffusion_name for the UNet, clip_name + clip_type for the text encoder, vae_name for the VAE. All three file lists match what ComfyUI's own core loaders (UNETLoader, CLIPLoader, VAELoader) expect, which means the dropdowns are populated by the same folders you already use. If you've been running Flux or another split-model pipeline with three separate loader nodes, this collapses them.

The inputs and outputs

Inputs: diffusion_name, clip_name, clip_type (the CLIP architecture - Flux, SD3, Hunyuan, etc. - pick what matches your model), vae_name, then the sampler settings (sampler, scheduler, cfg, steps), the latent settings (aspect_ratio, width, height, aspect_mode, batch_size, megapixels, multiple), seed, the SpeedLoRA trio, and the same save-prep inputs as AUNInputs (MainFolder, ManualName, name_mode, prefix, date_format, crop, words).

Outputs: MODEL, CLIP, VAE, model name, sampler, scheduler, cfg, steps, latent, width, height, seed, and the filename family - MainFolder, Filename, prefix, date format, batch size, name mode.

The one input that trips people is clip_type. It must match the CLIP file you loaded - a Flux CLIP with the SD3 architecture selected will fail or misbehave. The tooltip tells you exactly what it's for; believe it.

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 ModuleNotFoundError: cv2? pip install -r custom_nodes/AUN-ComfyUI-Nodes/requirements.txt.

Common issues

Beyond clip_type mismatches, the same "legacy directionally" caveat from AUNInputs applies: this full-style node carries the older save-prep outputs, and the pack recommends the Basic family for new workflows. If a dropdown shows <no diffusion models found>, you're looking at a fresh install where the diffusion/CLIP/VAE folders are empty - the node reads whatever's in your models folders, so it's a "models not downloaded yet" signal, not a broken node.