Visualize Context Options (K.Adv.) 🎭🅐🅓

This is the Visualize Context Options family's answer to KSamplerAdvanced-style workflows - the ones where you're not running the full 0-to-steps denoise in one shot, but slicing sampling into a specific start_step/end_step range. That's a common shape in multi-stage AnimateDiff pipelines: a base pass, then a second sampler picking up partway through for refinement. If that's your setup, the plain "K." visualize node won't reflect it, because it only knows steps and denoise - this one takes the step range directly.

How it works

Same idea as its sibling: it doesn't touch pixels or run the motion module, it simulates the sampling schedule your settings would produce and renders which frames land in which context window at each step, as a chart. The difference is entirely in what schedule it's simulating - instead of a full run scaled by denoise, it reconstructs exactly the step slice a KSamplerAdvanced node with the same start_step/end_step would actually execute, so the preview matches a partial or staged run rather than a complete one.

The inputs and outputs that matter

• model (MODEL, required) - the model you're sampling with; the step schedule depends on it.
• sampler_name / scheduler (COMBO, required) - match these to your real advanced sampler node.
• context_opts (CONTEXT_OPTIONS, optional) - the plan you're checking. Leave unplugged to preview with no windowing applied.
• steps (INT, default 20) - the total step count your schedule is built against, same as the total steps field on a KSamplerAdvanced node.
• start_step (INT, default 0) and end_step (INT, default 20) - the slice of that total schedule you're actually previewing. This is the whole point of the "Adv." variant: set these to match the exact range your real advanced sampler node runs.
• visual_width (INT, default 1440) - cosmetic width of the output chart.
• latents_length (INT, default 32) - how many frames the simulation assumes; doesn't need to match an actual batch you have on hand.

Output is a single IMAGE, same as the rest of the family - feed it to Preview Image or Save Image.

How to install it

Through ComfyUI Manager: search AnimateDiff Evolved, confirm the author is Kosinkadink, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved

then restart. No motion module needed for this node specifically - it's a schedule-and-windowing simulator, not a generator - but you'll need one loaded for the actual pipeline you're validating.

Common issues & troubleshooting

The preview doesn't match your real run. The whole value of this node depends on steps, start_step, and end_step exactly mirroring the advanced sampler node you're actually going to use. If you change your real pipeline's step range and forget to update this node to match, you're looking at a stale preview - treat these fields as something you keep in sync, not a one-time setup.

You're not actually running a staged/partial pipeline. If your real generation is a single plain KSampler with a denoise value rather than a start/end step slice, use the sibling Visualize Context Options (K.) node instead - its steps/denoise pair maps more directly onto that setup than juggling start_step/end_step would.

You built the sigmas by hand. If your real pipeline skips sampler_name/scheduler entirely and uses a custom SIGMAS object, neither this node nor the plain "K." one will reconstruct it faithfully - that's what Visualize Context Options (S.Cus.) is for.

Windows look fine here but the real run still seams. Remember this is only simulating the scheduling logic, not running the actual motion module or your ControlNets. A clean-looking window chart tells you your Context Options plan is internally consistent; it doesn't guarantee the visual result, since content-level seams can come from the model or your prompt schedule too.