Adjust Weight [All◆Mult] 🎭🅐🅓

This is the blunt instrument of AnimateDiff-Evolved's Weight Adjust family: one multiplier, applied to every weight in the loaded motion module at once. If Scale/Effect multivals (the ones the README describes for controlling "motion amount and motion model influence") feel like turning down the model's output, all_MULT feels closer to turning down the model itself - it scales the actual weight tensors before inference runs.

What it's for

AnimateDiff-Evolved's Weight Adjust nodes exist for people who've hit the limits of the normal motion-strength controls and want to reach directly into the motion module's weights instead. all_MULT is the coarsest one: a single float that multiplies every weight category the module has - positional encoding, attention, everything else - by the same factor. It's the node to reach for when you just want "less motion module influence, uniformly" without caring which internal component is responsible, as opposed to the Indiv (Individual) variants that let you scale positional-encoding weights, attention weights, and the rest independently.

Because it's a blanket multiplier on the model's actual weights rather than a runtime scale on its output, expect it to behave differently than turning down scale_multival on your Apply/Loader node - dial it too far from 1.0 in either direction and you're not just getting "less motion," you're changing what the module computes internally, which can degrade coherence rather than gently reduce it.

Inputs and output

• all_MULT (float, default 1, range 0–2) - the multiplier. 1 is unchanged. Below 1 dampens the module's influence; above 1 amplifies it, though the useful range in practice is much narrower than the full 0–2 the node allows.
• print_adjustment (bool, off by default) - logs the computed value. Worth turning on the first time so you're not tuning blind.
• prev_weight_adjust (optional, WEIGHT_ADJUST) - chain this onto a previous weight adjustment to stack effects, including combining it with one of the Indiv nodes if you want a global scale plus fine per-category tweaks.
• Output: WEIGHT_ADJUST - a settings object, not a visible result. 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. Or manually:

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

then restart. No extra downloads for this node specifically - it's pure settings - but you need a motion module (mm_sd_v15_v2, v3_sd15_mm, etc.) in ComfyUI/models/animatediff_models for it to attach to.

Common issues

You wanted "less motion" and got broken frames instead. This is the most common trap: all_MULT scaling the raw weights isn't a linear "less movement" dial the way effect_multival is. Push it too far off 1.0 and you can get artifacting rather than subtlety. If you just want to tone down how much the motion module influences the output without touching its internals, scale_multival/effect_multival on your Apply node is the more predictable place to start; reach for weight scaling only once that hasn't gotten you where you want.

Nothing changed. Same failure mode as the other settings nodes in this family - the output is a WEIGHT_ADJUST object that does nothing until it's actually wired into your model settings node. Turn on print_adjustment and confirm the value is reaching where you think it is.

You actually want per-component control. If uniform scaling isn't getting the specific effect you're after - say, you want less positional "drift" but the same attention strength - this is the wrong node. Reach for ADE_AdjustWeightIndivMult or the attention-specific variant instead, which split the multiplier by weight category.

KB gap: we don't have documented guidance on what multiplier ranges are actually usable before quality degrades noticeably, or how this differs in practice from scaling via effect_multival - that would be worth capturing from real user reports next time this pack comes up.