TT KSampler

This one is refreshingly boring. TT KSampler is just a KSampler - model, positive, negative, latent, seed, steps, cfg, sampler_name, scheduler in, one LATENT out. No context sockets, no proprietary types, no two-stage cleverness. If you've used stock ComfyUI's KSampler, you already know how to use this node, because it does the same job with the same parameter names.

It ships in TenserTensor, a pack whose entire personality is "combine chains into fewer, cleaner nodes." A plain sampler is the odd one out - which tells you something about why it exists: it's the reliable baseline of the pack's sampler family, the one that matches stock behavior so you can swap it in and out without surprises.

The inputs, for completeness

• seed - reproducibility. Same seed + same everything = same image.
• steps (default 25) - how many denoising steps.
• cfg (default 1.5) - the guidance scale. Note that default: it's tuned for modern flow-matching, guidance-distilled models like FLUX, not the old SDXL habit of cfg 7. If you feed it an SDXL model, 1.5 will look flat - raise it. The KB's rule of thumb: DPM++ 2M with a Karras scheduler is the safe default for SD 1.5/SDXL, but for flow-matching models you want Euler on a simple schedule.
• sampler_name (44 choices) and scheduler (9 choices) - the full ComfyUI list, so anything your model likes is available.

Under the hood it wraps comfy.sample.sample with sensible defaults: full noise, full denoise, disable_noise off. Output is a LATENT ready for a VAE decode - this pack's TT VAE Decode (Context) if you're in a context flow, or stock VAEDecode anywhere.

The honest take

If you're on stock ComfyUI and happy, there's no reason to install a whole pack for this node - the built-in KSampler is identical. TT KSampler earns its place if you're already building in the TenserTensor style and want every sampler in the graph to come from one family with matching defaults and the pack's latent-preview callback. It's also a fine teaching node: no extra knobs to trip over, so it's a clean first sampler to understand how latents flow through a graph.

Install and the V1 note

cd ComfyUI/custom_nodes
git clone https://github.com/tenser-tensor/ComfyUI-TenserTensor

or ComfyUI Manager → "TenserTensor" → install → restart. Deps: gguf and kornia (the pack's requirements.txt), installed automatically via Manager.

Standard caveat for this pack: it's a legacy V1 node, migrated to Deprecated/ after the pack's move to ComfyUI's API V3, and slated for removal in a future major release. For a node whose job is matching stock behavior, that's an easy loss to absorb - stock KSampler does the same thing, so when this one goes away, the fix is a two-second node swap. One genuinely useful habit regardless of which sampler you use: keep seed wired to a fixed value or a seed node while you tune, then randomize once the image is good.