CFG Extras◆RescaleCFG [Multival] 🎭🅐🅓

If you've ever pushed CFG up to fight weak prompt adherence and watched your image blow out - too much contrast, colors clipping toward pure white or pure black - that's a known, well-understood artifact of classifier-free guidance at high scales, not a setting you did wrong. RescaleCFG is the fix: it rescales the guided prediction back toward the statistical range of the unguided one, keeping the extra push CFG gives you without the exposure blowout. This node wires that technique into AnimateDiff-Evolved's Sample Settings as a CFG Extra.

Why this exists

Classifier-free guidance extrapolates away from the unconditional prediction toward the conditional one, and the further you push that extrapolation (higher CFG), the more the result's variance grows beyond what the model was actually trained to produce - that's the mechanical cause of high-CFG overexposure. RescaleCFG measures that variance drift and pulls the final result back down to roughly match the conditional prediction's own statistics, so you keep the structural benefit of a strong CFG push without the washed-out look. It's not unique to AnimateDiff - ComfyUI has a native RescaleCFG node for exactly this reason, and other SD-based ecosystems reach for the same fix (Illustrious workflows, for instance, pair it with CFG++ samplers specifically to counter oversaturation). This node is AnimateDiff-Evolved's version, scoped to fit into its own Sample Settings/CFG Extras chain rather than ComfyUI's general sampling pipeline.

Inputs and output

• mult_multival (MULTIVAL, required) - how strongly the rescale is applied. 0 is effectively off; higher values pull harder toward the unguided prediction's range. Being a MULTIVAL, this accepts a single float, a list, or a mask, so it's schedulable across the clip the same way scale/effect multivals are elsewhere in the pack.
• prev_extras (optional, CFG_EXTRAS) - chain onto another CFG Extras node, such as ADE_CFGExtrasPAG, to combine multiple guidance adjustments.
• Output: CFG_EXTRAS - wire into your Sample Settings node so it reaches the sampler.

Installing the pack

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

restart. No separate model download for this node - it's a sampling-time adjustment - but the rest of your AnimateDiff graph still needs a motion module (mm_sd_v15_v2, v3_sd15_mm, etc.) in ComfyUI/models/animatediff_models.

Common issues

Output goes flat/muted instead of just less blown-out. Rescale too aggressively and you can pull the result too far toward the unguided prediction's dynamic range, losing the punch you raised CFG for in the first place. This is a balancing act, not a switch - start moderate and adjust from there rather than maxing it out.

You're not actually running high CFG. RescaleCFG solves a specific problem - overexposure from a strong CFG push. If your CFG is already modest and your images look fine, adding this node isn't going to improve anything; it exists to counteract a specific artifact, not to generally "improve" output.

No visible effect. Confirm CFG_EXTRAS is actually wired into your Sample Settings node - like every settings node in this pack, it's inert unless connected into the chain that reaches the sampler.

Stacking with PAG and not sure what's doing what. Both are CFG Extras chained via prev_extras; if results look off with both active, disable one at a time to isolate which adjustment is responsible before tuning further.