ComfyUI 0.43 brings Kling 4.0, PixVerse V6-Max, VEED Lipsync 2.0 and Mirelo SFX into the graph

ComfyUI v0.43.0 went out on Oct 26 at 8:40 PM MDT. For video, the release notes list new paid API (partner) nodes for Kling 4.0, PixVerse V6-Max, VEED Fabric 1.0 and Lipsync 2.0, Mirelo SFX, the ClarityAI Crystal Video Upscaler, FLUX 3 Video at 2K/4K, and HeyGen voice clone and TTS. On the open side, kijai added support for single-head LoRAs on MiniMax H3's PDD head banks, MOSS-Image is supported, and there's ACEScg EXR handling for anyone grading in a VFX pipeline, plus new Branch and Pad Image utility nodes.

Why Kling 4.0 matters here. The model reached third-party APIs last week (fal, Runway, Runware) with clips of up to 30 seconds with native audio, multiple keyframes and many references, according to those platforms' announcements. With the node, Kling can sit next to H3 or Wan in the same graph for shot-by-shot comparisons. Creator @maxxmalist's per-clip numbers put a Kling 4.0 shot at about $2.10 against $1.38 for Wan 3.0. Noble Goose's anime stress test rated Kling 4 a solid model, about level with MiniMax H3, but found Seedance 2.5 better at staging when the 15-second, dialogue-heavy shots got demanding.

The audio and finishing nodes may matter more than the headline. Lipsync 2.0 and Mirelo SFX let one workflow carry a silent generation through to a lip-synced clip with sound effects, and Crystal handles the upscale at the end. These are API nodes, so each run costs partner credits; check prices in the node before you batch anything.

Our take: update ComfyUI if you compare models per shot. The cheapest way to try Kling 4 is a handful of hero shots in a graph you already use, with H3 or Wan doing the drafts. Winnougan's guide is also worth a look if you want custom H3 LoRAs: per the video, a 30-image MOSS/H3 character LoRA trained in under three hours and never went above about 9.5 GB of VRAM.