Rebel KSampler ⚙️

Rebel KSampler ⚙️ is the core KSampler node with an animated body - which is to say it is the node where your entire image actually gets made, wearing the pack's Matrix rain (or a ball bouncing, or a car drifting, you do you). Every dial that matters on the stock sampler is here with the same tooltips, same defaults, same outputs. If you swap this in for your existing KSampler, your images come out identical. What you get instead is the satisfaction of watching a cyberpunk terminal backdrop behind the sliders while your generation runs. There is no performance benefit. There is no quality benefit. It's a dashboard for the part of the graph that deserves one, because the KSampler is the one node doing all the heavy lifting.

How it works

Like every wrapper in the pack, it asks ComfyUI's native KSampler for its real input definitions, adds theme and color_scheme, and delegates the actual sampling call straight through to the core node. The sampling math - noise generation, step-by-step denoising, scheduler - is literally ComfyUI's own code running underneath. The animated backdrop is drawn by the pack's JavaScript in the node's background layer and costs no sampling time and no VRAM; the author's own pitch is accurate, "they're pretty much overlays."

The inputs that matter

Seven of them do real work, and they're the ones you already know from stock ComfyUI:

• model - the diffusion model to denoise with, wired from your checkpoint or UNET loader.
• positive / negative - the CONDITIONING from your prompt encode. Positive is what you want, negative is what you want kept out.
• latent_image - the latent to start from, usually an Empty Latent or a VAE-encoded image.
• seed - noise seed; remember the control_after_generate dropdown next to it (fixed, randomize, increment, decrement). The classic beginner trap: by default the control fires after the run, so the seed shown is the one that ran last, not the one you just used. Set it to fixed when you find an image you like, and consider flipping the global "widget control mode" to Before once.
• steps - denoising steps, default 20.
• cfg - classifier-free guidance scale, default 8. Higher clings harder to the prompt; too high and quality falls apart.
• sampler_name / scheduler - the algorithm and noise schedule. The dropdowns carry ComfyUI's full current list, so you're not giving up any sampler options by using the Rebel version.
• denoise - 1.0 means full generation from noise; lower values preserve structure, which is how img2img-style refinement works.

The output is a single LATENT, feeding your VAE decode.

Installing it

The pack is one clone, zero extra dependencies:

cd ComfyUI/custom_nodes
git clone https://github.com/RealRebelAI/Rebels_Animated_Nodes.git

Restart ComfyUI, then hard refresh the browser (Ctrl + F5 / Cmd + Shift + R). Without that refresh, ComfyUI's cached JavaScript means you'll see the node but the animation won't draw - it's the pack's #1 support question. Find it under Rebel AI → Animated Core.

Where people get burned

• Same settings, different image than your stock KSampler gave you? That shouldn't happen - the wrapper passes everything through. If you're seeing different results, check the seed, not the node; a randomize control_after_generate will happily roll new seeds every run on either node.
• Animations flat after upgrading ComfyUI. This pack renders on the legacy LiteGraph canvas; on the Nodes 2.0 frontend the background animation tends to vanish even though sampling works fine. Keep the classic canvas if the aesthetic matters.
• Expecting new sampling features. You won't find any - this is the stock sampler list, dressed up. If you want fancier sampling, the ecosystem has dedicated packs for that; this node is the familiar one in a fancy box.

It's the node you'll stare at longest, so it's the node where the pack's whole premise pays off most. If the gray box has been depressing you through long batch runs, this is a genuinely pleasant swap. Just don't expect it to change a single pixel.