🧩AGSoft KSampler

ComfyUI's stock KSampler is fine until you want to know what you actually ran. Seed, steps, CFG, sampler, scheduler, denoise - they're all sitting in widgets, but if you want them in a log, a filename, or a workflow summary, you have to wire each one out by hand. AGSoft KSampler is a drop-in KSampler that also hands you every parameter as strings and JSON. If you've ever stared at a folder of outputs wondering which settings made which image, this is the node that answers.

What you get

The sampling side is stock behavior: model, positive/negative conditioning, latent_image, seed (with control-after-generate), steps, cfg, sampler_name, scheduler, denoise - all standard, and it uses ComfyUI's own common_ksampler under the hood, so results match what you'd get from a normal sampler. The deltas are the outputs.

• latent - the denoised latent, same as stock.
• image - a decoded image, if you connect the optional vae. That's a genuinely useful convenience: sampler + VAE decode in one node, no separate VAE Decode needed.
• Then the receipt: seed_str, steps_str, cfg_str, sampler_str, scheduler_str, denoise_str - every parameter as a string you can splice into a filename, a text label, or a Save Image caption.
• params_json - the whole lot as a JSON document, ready for structured logging or auto-generated workflow notes.

The design intent is obvious from the output list: this is a node for people who treat generation as an experiment and want provenance attached. It's the difference between "I ran something" and "I ran euler_ancestral, seed 42, CFG 7.5, 28 steps."

One honest caveat

If you don't connect the VAE, the image output is a 1×1 dummy tensor - the node deliberately avoids crashing and just hands back a placeholder. So if you wire image somewhere without a VAE connected, you'll get a black pixel, not a preview. Use the latent output plus a real VAE Decode in that case, or connect the VAE.

Installing it

Pack standard: ComfyUI Manager → search comfyui-AGSoft → install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft.git

No models, no dependencies beyond ComfyUI's core - it's a thin wrapper over the built-in sampler.

Common issues

The VAE dummy-tensor behavior above is the one that'll catch you. Other than that, it behaves exactly like a stock KSampler, so any sampler troubleshooting you already know applies. If you're comparing results against another KSampler in the same graph, remember seed still respects control-after-generate, so both nodes will change seed together when you hit queue. And for style-transfer or img2img use, denoise below 1.0 works as expected - the logging just records what you set, which is the point.