Trending Model:#1Inklingthinkingmachines⬇16kTrending Model:#2Ternary-Bonsai-27B-ggufprism-ml⬇432kTrending Model:#3Unlimited-OCRbaidu⬇2237kTrending Model:#4Bonsai-27B-ggufprism-ml⬇1405kTrending Model:#5GLM-5.2zai-org⬇545kTrending Model:#6Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUFDavidAU⬇63kTrending Model:#7Qwen3.6-35B-A3B-Uncensored-HauhauCS-AggressiveHauhauCS⬇1998kTrending Model:#8Qwythos-9B-Claude-Mythos-5-1M-GGUFempero-ai⬇2133kTrending Model:#9krea2-identity-editconradlocke⬇0kTrending Model:#10Laguna-S-2.1poolside⬇3kTrending Model:#1Inklingthinkingmachines⬇16kTrending Model:#2Ternary-Bonsai-27B-ggufprism-ml⬇432kTrending Model:#3Unlimited-OCRbaidu⬇2237kTrending Model:#4Bonsai-27B-ggufprism-ml⬇1405kTrending Model:#5GLM-5.2zai-org⬇545kTrending Model:#6Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUFDavidAU⬇63kTrending Model:#7Qwen3.6-35B-A3B-Uncensored-HauhauCS-AggressiveHauhauCS⬇1998kTrending Model:#8Qwythos-9B-Claude-Mythos-5-1M-GGUFempero-ai⬇2133kTrending Model:#9krea2-identity-editconradlocke⬇0kTrending Model:#10Laguna-S-2.1poolside⬇3k

Comfyui-lrw-nodes Brings Geodesic Guidance for Smooth AI Video Frames

By vramkickedin | | 2 min read
A complex geodesic wireframe sphere that transforms into elegant fluid ribbon curves.

Comfyui-lrw-nodes is a newly released set of custom nodes for ComfyUI that brings latent-space geometry tools into local AI image and video workflows. It lets users experiment with geodesic interpolation, curvature analysis, and parallel transport without needing to understand the underlying math. The standout feature is practical guidance for WAN2.2 First-Last Frame video generation, making intermediate frames smoother when the start and end images differ a lot.

Developer Lajjadred built the node pack on top of the latent-riemannian-world library to make advanced latent manipulation accessible inside ComfyUI. The project addresses a common problem with FLF videos: abrupt or unstable motion when the two input frames have large changes in pose or camera distance. By keeping the workflow VRAM-safe, it runs on consumer GPUs from 12 GB upward, which is important for local AI users.

Practical WAN2.2 guidance with geodesic blending

Key Features
  • Adds geodesic interpolation and curvature analysis nodes.
  • VRAM-safe WAN2.2 guidance for consumer GPUs.
  • Softly blends geodesic keyframes into video latent.
  • Parallel transport for moving style vectors.
  • Includes SLERP interpolation for baseline tests.
  • Pre-made example workflows to get started fast.
  • Depends on latent-riemannian-world version 0.3.0.
  • Works with UnetLoaderGGUF and VHS Video Combine.

Video creators can use the geodesic blending to improve animation smoothness when generating motion between two drastically different stills. Because everything stays on your own machine, no data ever leaves your workstation, which suits sensitive client work or personal projects.

What to know before you start

The WAN2.2 integration is intentionally pragmatic, avoiding heavy VAE decoder Jacobian calculations that would choke most consumer GPUs. New users should start with low blend_strength settings (0.10 to 0.15) and enable normalize_guide to prevent weakening the final frame arrival. The workflow never replaces the intended last frame with a midpoint—it only adds a geodesic guide before the sampler, preserving continuity.

"The recommended workflow does not replace the last frame with a midpoint. Instead, it keeps the actual last frame in WAN FLF and blends LRW guidance into the video latent before sampling." Source: GitHub

Project Links