Context Extras◆NaiveReuse 🎭🅐🅓

NaiveReuse is the second of AnimateDiff-Evolved's two "novel cross-context consistency techniques," and it earns the "naive" in its name honestly: where ContextRef injects the reference window's actual attention features into later windows, NaiveReuse does something much blunter - it blends the reference window's noise statistics into later windows' starting point. No attention internals, no style-transfer math, just reusing what the earlier window already settled on as a nudge for what comes next. Cheaper to compute, less precise, and - going by the defaults - meant to be used briefly rather than throughout the whole render.

What it does

Like ContextRef's root node, everything here is optional, so plugging this in with nothing configured still gives you working defaults. strength_multival is the overall dial - a flat float, per-frame list, or mask, same Multival pattern as everywhere else in this pack. naivereuse_kf takes a keyframe chain if you want the strength to change across the sampling process instead of staying flat. weighted_mean (default 0.95, 0–1) is the actual blend weight - how strongly a later window's starting noise gets pulled toward the reference's, versus staying its own. start_percent and end_percent (0 and, notably, 0.15 by default) bound when this applies in the denoising process, and that 15% default is a real signal about how the technique is meant to be used: broad composition, color, and structure get decided early in diffusion sampling, so that's when a cheap noise-level nudge toward a reference actually helps. Run it deep into the later steps and you'd be fighting fine-detail refinement instead of assisting composition, which is presumably why the pack doesn't default to that. prev_extras chains this with a ContextRef node so both effects can run together - CONTEXT_EXTRAS is a linked list, not a single slot.

Output is CONTEXT_EXTRAS, and - same as ContextRef - it does nothing on its own. It has to be routed through Set Context Extras, wired alongside your Context Options node, before the sampler ever sees it.

NaiveReuse vs. ContextRef

Reach for NaiveReuse when you want a lighter touch, or when ContextRef's attention-injection machinery is more than the problem calls for - a subtle color or composition drift across a long render, rather than a full identity lock. Reach for ContextRef when the drift is specifically about identity or style holding together, which needs the deeper attention-level intervention. Nothing stops you chaining both via prev_extras if a render needs both kinds of help.

Installing it

ComfyUI Manager: search AnimateDiff Evolved, author Kosinkadink, install, restart. Manually: git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved into custom_nodes, restart. As with ContextRef, this node configures cross-window behavior - it doesn't replace the motion module you still need to download per the README's Model Setup section.

Where you're on your own

Same honest note as ContextRef: searching the reddit archive for the exact phrase "NaiveReuse" turns up zero threads. This is not a feature with an established community playbook - you're testing on your own render, not following someone else's tuned settings. The one grounded piece of advice worth keeping in mind: if you don't see any effect, check end_percent isn't set so low it's only running for a step or two, and confirm the chain actually reaches Set Context Extras.