comfy.hebolide.com

May Wrapped

CN
2026-06-02 02:03:19

If you missed it, here's a quick recap.

Throughout May, we incorporated 11 fresh models spanning visual media, pictures, three-dimensional spaces, audio signals, and multimodal systems. Discover the recent updates below.

Krea 2 — Krea AI · Image & Style Transfer

  • Krea revealed its foundational model first, immediately accessible as a Partner Node. While many image systems emphasize content composition, Krea 2 distinguishes itself through stylistic execution—handling aesthetic influences, inspiration boards, and creative variations across drawings, animations, lifelike imagery, and further genres.

Void — Netflix · Video Object Removal

  • Netflix released VOID under open-source licensing for video-based subject elimination. Traditional repair methods target pixel removal, but VOID erases both the element and its consequential traces—like shadows or reflections. Supporting Apache 2.0, it integrates natively.

Tripo 3.1 — Tripo AI · 3D Generation

  • Integrates generative processing from text, visuals, or multiple angles into one module. Combined with TripoSplat, it creates an end-to-end pathway for converting pictures into 3D-Gaussian representations.

Luma UNI-1 — Luma AI · Image Editing

  • Not structured as a diffusion solution, Uni-1 utilizes decoding-focused autoregressive transformers for prompt analysis before creation. Enhanced reference precision enables editing through Create/Modify modes with up to nine visual inputs. Among this year's most innovative designs.

Claude — Anthropic · Multimodal

  • Anthropic's system became available directly in workflows. Handle linguistic tasks, pipeline logic, or multimodal comprehension anywhere using its AI intelligence.

OpenRouter · Text

  • Single-node entry to twenty-plus large language models. Direct requests to optimal algorithms without exiting processes.

Gemma 4 — Google DeepMind · Multimodal

  • Google's premier accessible framework yet: handles text, images, sound, and video. Operates on mobile or lone GPU systems, featuring a 31B variant ranked third on Arena's rankings with Apache 2.0 and vast context capacity.

HidDream-O1-Image — Hidream.ai · Image

  • Reasoning-oriented visual creation using an open-source Pixel-level Unified Transformer (UiT). Excels with intricate prompts where rivals falter, avoiding separate encoders.

Stable Audio 3 — Stability AI · Audio & SFX

  • Generates audio tracks, sound effects, and production-ready output from textual cues. Your sonic toolkit now integrated.

BiRefNet — CAAI AIR · Background Removal

  • High-definition foreground segmentation. Naturally complements VOID—BiRefNet processes static frames while VOID manages movement.

MoGe — Microsoft · 3D Geometry & Depth

  • Derives comprehensive 3D structures—spatial coordinates, elevation metrics, surface orientations, lens parameters—from singular pictures. A CVPR '25 Oral presentation addition.

ComfyHub is gaining momentum

  • Exceeded 500+ stored workflows

  • Significant likelihood that another user developed the setup you planned

Anticipate more developments in June!

Recommend