ComfyUI ecosystem updates — open-source model releases, custom nodes, workflows, and tools for image, video, and audio generation.
Text to imageComfyUI nodes that build a person once across 44 fields, then generate a full series of photos: framing, pose, expression and aspect ratio vary while the person stays the same.
Character animationAn addon for ComfyUI-H3-Motion-Context that auto-chains MiniMax H3 lip-sync clips: audio is split, motion and audio carry through latent context, and clips stitch into one MP4.
Character animationPolyU, ByteDance, and AMD release Avatar-Forever, a real-time audio-driven avatar model built on LTX 2.3 22B with ForeverCache streaming: unbounded generation at 27.2 FPS.
Image to 3DMeta's SAM 3D Body human mesh recovery is now built into ComfyUI core: video pose tracking, facial expression driving, and GLB/BVH export via new SAM3DBody nodes.
SAM-License
Image to 3DComfyUI core now supports TRELLIS.2 and Pixal3D image-to-3D: shape and texture stages, MoGe depth, and mesh post-processing nodes with official Comfy-Org weights.
MIT
Text to imagelylogummy releases Anima-3.8B Pro52 / Qwen3.5 Edition, a 52-block anime DiT that pairs the Anima visual style with Qwen3.5 4B language understanding, with dedicated ComfyUI nodes.
Text to videoA community graft transplants Z-Image's spatial attention onto MiniMax-H3 via q_norm rescaling: richer sets and textures with flat detail across joins, drop-in ComfyUI checkpoints.
minimax-h3-community-license
Text to imageA community distillation LoRA cuts Krea 2 Turbo from 8 steps to 4 with texture at teacher parity and a prompt-aware critic. Plain LoRA weights, no custom nodes needed.
Image to videoLightx2v releases MiniMax H3 Turbo-SLA, a 4-step FL2V distillation with 85% sparse attention, about 2.5x faster on RTX 5090, in LightX2V and ComfyUI formats.
Image editingiamkaikai trains the H3 image decoder for 500K steps, producing a single-frame VAE checkpoint that decodes one image per H3 latent slice with sharper, more coherent output.
Video editingQwen-Video-Edit repurposes Qwen-Image-Edit's DiT to edit Wan 2.1 video latents from text instructions, with official ComfyUI custom nodes and 360P/480P/720P checkpoints.
MIT
Text to imageLodestones ships Kroma v0.3 for Krea 2: a non-distilled full base checkpoint, a turbo variant, and a community LoRA pack that unlocks prompt following fixes.
EcosystemComfy and MiniMax challenge you to create a video where sound and motion are inseparable using MiniMax H3 in ComfyUI, with RTX 5090 and RTX 5060 Ti prizes.
Text to imageSenseTime ships the full SenseNova-U1.5-8B-MoT: OPD-distilled experts, better 4K quality, editing preservation, and text rendering, all runnable in ComfyUI.
Text to videoRAVEN (Imperial College London) turns MiniMax H3 into a real-time streaming video generator via a 4-NFE LoRA adapter with official ComfyUI nodes for chunk-by-chunk output.
minimax-h3-community-license-agreement
Text to videoLightx2v's 8B prompt-rewriter LoRA for MiniMax H3 now handles image-conditioned rewriting, and Nynxz packages it as ComfyUI-NynxzH3 v0.0.2 with a drop-in node.
Video editingTencent releases CoinVE-Edit, a 22B compositional video editing model on Wan2.1-T2V-14B and Qwen3-VL-8B that applies 2-5 region-aware edits in a single pass.
Apache-2.0
Video editingHud224 ships Cadence Retime, a ComfyUI node pack that linearizes uneven H3 motion and applies Twixtor-style speed ramps with audio sync, without After Effects.
Text to videoLBH-123 releases a 3D-conv neural latent upscaler for MiniMax H3 with ComfyUI custom nodes, upscaling 24-channel video latents in-place to speed up high-resolution generation.
Text to videoNVIDIA Sol Engine runs MiniMax H3 as a 4-step draft plus 3 LTX refinement steps: 6.85 s for 5-second 768p video, up to 27.7x faster than SGLang on one GB200.