ComfyUI-LatentClamp

A node for ComfyUI that modifies the values in the samples input that fall outside of a min and max range using a multiplier.

Additional controls are available for experimental purposes.

In img2img operations, latent clamping has been observed to improve prompt adherence. Issues with image fidelity resulting from the clamp can be rectified with a quick second diffusion pass.

Inputs

• min: Values in samples below this value will be multiplied by outside_multiplier.
• max: Values in samples above this value will be multiplied by outside_multiplier.
• inside_multiplier: Values in samples between min and max will be multiplied by this value.
• outside_multiplier: Values in samples outside of min and max will be multiplied by this value.
• extra_noise: Add compensatory noise to the values in samples.