Skip to content

1.10.0

Latest

Choose a tag to compare

@vslinx vslinx released this 15 Jun 20:41
  • added new Forward/Bypass-Mute on State (Any)-Node in the vsLinx/utility group. Forwards any value while mirroring the bypass/mute state of the node connected to its trigger input onto the directly connected downstream node(s) (bypass → bypass, mute → mute, normal → normal). Includes an Ignore subgraph boundary toggle to follow the trigger across subgraph boundaries until a real node, and a Mirror this node's own bypass/mute toggle to also propagate this node's own state downstream.
  • added new VAE Decode (Batched) and VAE Decode Tiled (Batched) nodes in the vsLinx/latent group. They work exactly like ComfyUI's built-in VAE Decode / VAE Decode (Tiled) but add a batch_size field that controls how many latents are decoded by the VAE at once (default 1). Decoding fewer at a time lowers peak VRAM and can speed up generation; values >= the batch size behave identically to the built-in nodes.
  • added new (Impact-Pack) Interactive Detailer-Node in the vsLinx/detailer group. A FaceDetailer clone that, instead of a wildcard field, pauses the workflow and pops up a dialog to enter a prompt per detected segment (empty = base prompt, [CONCAT] to append, [SKIP] to leave untouched), then details each segment with its own prompt. Requires ComfyUI-Impact-Pack.
  • added lanczos as an additional upscale_method option to the Upscale by Factor (With Model)-Node. Lanczos gives the sharpest, highest-quality resize but runs on the CPU (via PIL), so it is somewhat slower than the other methods.