KSamplerAdvanced (Texturaizer)

If the Texturaizer pack has a heart, it's this node. KSamplerAdvanced (Texturaizer) is a full re-implementation of ComfyUI's Advanced KSampler with two extras bolted on: an A1111-style "GPU" noise mode and a batch seed mode - and it's the natural landing spot for every output of Get Sampler Data. You'll probably never touch the stock KSampler again inside this pipeline.

Context: Texturaizer is a paid Blender addon by Luke Kratsios (LatentSpaceDirective); these nodes are the free ComfyUI companion half that turns Blender's exported JSON into actual sampling.

The inputs

Same shape as stock KSampler Advanced - model, add_noise, noise_seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, start_at_step, end_at_step, return_with_leftover_noise - plus the two it adds:

• noise_mode - GPU(=A1111) or CPU. The label is the tell: A1111 generates noise on the GPU, which is the behavior most people coming from Stable Diffusion WebUI expect. CPU is the default ComfyUI-style approach and produces identical noise for the same seed, just via a different generator.
• batch_seed_mode - incremental or comfy. With incremental, each image in a batch gets seed + i, so every slot in the batch differs; with comfy, the whole batch shares the seed. This is the knob people use to make multi-image batches reproducible instead of all-identical or all-different-by-accident.

The scheduler list is also slightly extended: stock schedulers plus the AYS SDXL, AYS SD1, AYS SVD, and GITS[coeff=1.2] presets (Align Your Steps - tuned sigma schedules that shave steps at the same quality - and the GITS coefficient variant).

How it actually works - and the dependency you must not skip

Here's where it stops being a clone. The sampling core routes through Impact Pack's RegionalSampler (impact_sampling()), which is how the pack supports regional/segmented prompting per object. If ComfyUI-Impact-Pack isn't installed, the node raises:

[ERROR] You need to install 'ComfyUI-Impact-Pack'

That's not optional fanciness - it's a hard requirement. Install Impact Pack via ComfyUI Manager before you wire this node, or your first run fails with that exact message. Impact Pack is one of the ecosystem's most-installed packs anyway (it's where FaceDetailer and SAM live), so you'll probably want it regardless.

It also inherits the usual KSampler Advanced semantics: start_at_step/end_at_step give you the "skip the beginning / stop before the end" control that makes img2img and regional workflows tick, and return_with_leftover_noise controls whether the last step leaves noise (needed when you're going to feed the latent into another sampler).

Wiring

In a Texturaizer graph this node is almost always fed by Get Sampler Data: the eleven typed outputs drop straight onto the matching inputs - seed→noise_seed, cfg→cfg, sampler→sampler_name, scheduler→scheduler, steps→steps, and so on. use empty latent from the data node maps to whether you feed it a fresh Empty Latent or a loaded image.

Install

The pack plus Impact Pack:

cd ComfyUI/custom_nodes
git clone https://github.com/LatentSpaceDirective/ComfyUI-Texturaizer

Or: ComfyUI Manager → search "Texturaizer" → Install, then search "Impact Pack" → Install, restart ComfyUI. The pack's own requirements.txt is just stock ComfyUI packages (torch, numpy, scipy, Pillow, torchvision) - no extra model downloads. If the sampler behaves differently from stock ComfyUI on identical settings, the Impact Pack routing is the reason; that's a feature here (it's what powers segmented per-region sampling), not a bug.