SimpleSamplerVAE

SimpleSamplerVAE is the pack's "whole pipeline in one box" node: you give it a model, a clip, a VAE name, raw prompt text, and dimensions - and it internally encodes the prompts with wildcard expansion, creates the empty latent, samples, and decodes back to an IMAGE. No CLIPTextEncode, no EmptyLatentImage, no KSampler, no VAEDecode. The entire chain collapses into one node.

Where this shines: keeping a workflow dramatically simpler than the standard graph while still getting wildcards. The stock ComfyUI way of doing the same thing is six nodes. Here it's one, and because the positive and negative are plain multiline strings, they get wildcard expansion for free - same __name__ / {a|b|c} syntax as the pack's encoders. If you're prototyping or want a minimal reproducible workflow to hand to a friend, this is the "fewest moving parts" answer.

Inputs and outputs

• model (MODEL), clip (CLIP) - your checkpoint.
• vae_name - a dropdown of the VAEs in your models/vae folder; this is the difference from the plain SimpleSampler, which takes an already-loaded VAE object instead.
• positive / negative (multiline STRING) - wildcard-enabled prompt text.
• width / height (default 512, in 64 steps), batch_size (default 1) - the generated latent's dimensions.
• seed, steps (20), cfg (3.0), sampler_name, scheduler, denoise (1.0) - the sampling settings, identical in spirit to a KSampler.

Output: one IMAGE, straight out of the VAE - wire it to a save node. The latent is generated internally, so there's no latent passthrough; treat this as a self-contained image generator, not a building block you can split.

Install and honest caveats

Clone and restart; ComfyUI Manager finds it by searching "Lilly":

cd ComfyUI/custom_nodes
git clone https://github.com/lilly1987/ComfyUI_node_Lilly

Restart after cloning. No models to download; rich and chardet auto-install on first load.

Honest caveats: the default CFG of 3.0 is SDXL-flavored; on SD 1.5 models you'll usually want it higher, and on guidance-distilled models lower - check what your checkpoint wants. And the wildcard-caching gotcha applies: ComfyUI caches unchanged inputs, so to re-roll a wildcard you need to change the text or seed. Finally, this is a 2023 pack that's unmaintained - the author's README points people to Impact Pack now. It still runs, but it's not where you should invest a new workflow.