Adjust Weight [Indiv◆Mult] 🎭🅐🅓

The middle ground in AnimateDiff-Evolved's Weight Adjust family: not as blunt as ADE_AdjustWeightAllMult (one multiplier for everything), not as fine as ADE_AdjustWeightIndivAttnMult (which splits attention into Q/K/V and output weight/bias). This one scales three categories independently - positional encoding, attention as a whole, and everything else - which for most tuning is the level of control that's actually useful.

Why split into three

The motion module's weights aren't a single homogeneous thing. Positional-encoding weights govern how the model reasons about where each frame sits in time; attention weights govern how frames relate to each other; "other" is the rest of the module. A single global multiplier moves all three together, which is fine as a coarse "less/more motion module influence" knob but can't isolate, say, "I want less temporal drift but the same amount of attention strength." This node gives you that isolation without dropping all the way into per-projection attention internals, which is usually more granularity than you need.

Inputs and output

• pe_MULT (float, default 1, range 0–2) - scales positional-encoding weights.
• attn_MULT (float, default 1, range 0–2) - scales attention weights as a whole.
• other_MULT (float, default 1, range 0–2) - scales everything outside PE and attention.
• print_adjustment (bool, off by default) - prints the computed values, useful to confirm the node is doing what you expect before you judge by eye on a full render.
• prev_weight_adjust (optional, WEIGHT_ADJUST) - chain onto a prior weight adjustment to stack effects.
• Output: WEIGHT_ADJUST - a settings object, invisible on its own. Wire it into whichever node in your graph collects per-model AnimateDiff settings, which attaches to your motion model loader.

Installing the pack

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

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

restart. No model file needed for this node specifically, but a motion module (mm_sd_v15_v2, v3_sd15_mm, and the rest) has to be downloaded into ComfyUI/models/animatediff_models for it to have something to attach to.

Common issues

You're trying to fix "too much motion" here first. Before touching weight multipliers, try scale_multival/effect_multival on your Apply/Loader node - those scale the motion module's runtime influence rather than its actual weights, and are the more predictable, more commonly used control. Weight scaling is for when that alone isn't giving you the specific effect you want, and you have a reason to believe one weight category (rather than overall strength) is the issue.

Pushing pe_MULT down to reduce jitter, but coherence gets worse instead. Positional encoding and motion aren't the same thing - scaling PE weights changes how the model tracks temporal position, not directly how much it moves. If your goal is literally "less motion," attn_MULT or the overall effect multival is the more direct lever; pe_MULT is more about stability/coherence over the sequence.

Nothing changed after adjusting values. Confirm the WEIGHT_ADJUST output is actually wired into your motion model's settings node - this family of nodes does nothing on its own until attached.

You need finer control than three buckets. If you've isolated that attention specifically is the problem and need to go deeper than a single attn_MULT, ADE_AdjustWeightIndivAttnMult splits attention into query/key/value/output weight/output bias.

KB gap: we don't have documented real-world guidance on typical useful ranges for these three multipliers, or which specific artifacts each one is known to fix - worth capturing from the pack's own docs or user reports the next time this pack comes up.