LoginSign Up
← Back to all sparks
C

ComfyUI

DESIGN
Velocity7.5

Node-based stable-diffusion UI for building custom image-generation workflows visually.

www.comfy.org ↗

ComfyUI absorbs every new open model within days — now including full song generation.

day-0-model-supportpartner-nodesopen-weightsmultimodallocal-inferencemusic-generation
◆Current state
ComfyUI is running a near-weekly model intake: LTX-2.5, Wan Animate 2 and MiniMax H3 arrived with day-0 open-weights support, while Seedance 2.5 and FLUX 3 came in as hosted Partner Nodes. The split is consistent — open weights get local optimization and hardware claims, closed models get a Partner Node wrapper. The runtime itself is barely changing; the node catalog is what grows.
◆Where it's heading
Two lanes are widening at once. The model lane is expanding past video and image into audio and now full music generation, making the graph a place to assemble any modality rather than a picture pipeline. The platform lane — Comfy for Teams, Comfy MCP, the education partnership — is where the actual product bets sit, turning a local tool into shared infrastructure with agent access.
◆Prediction
Expect the Partner Node roster to keep absorbing closed models on announcement day while the open-weights posts continue leading with local-hardware requirements, since that contrast is the pitch. Whether music generation becomes a supported lane or a one-off is not yet visible in the entries.

◆Recent moves

  1. 4d ago

    The 8 Best AI Creative Workflow Platforms in 2026

    A ranked listicle comparing AI creative workflow platforms, published to the same feed as ComfyUI's release notes. It is search-capture content rather than a release, and says nothing about what ComfyUI itself can now run.

    View source ↗
  2. 5d ago

    MiniMax Music 3: State of the Art Open Weight Music Generation

    ⚡ SPARK

    MiniMax Music 3 brings lyric-driven song generation into a graph that until now produced images and video. It is the first entry in this window where the addition is a new output modality rather than another engine for one ComfyUI already had.

    View source ↗
  3. 7d ago

    LTX-2.5 Day-0 Support in ComfyUI

    LTX-2.5 arrives on day one with multi-shot generation and a fine-tunable checkpoint. Notable for the fine-tuning angle, but day-0 model support is ComfyUI's standing cadence rather than a change in direction.

    View source ↗
  4. 11d ago

    Wan Animate 2 is now available in ComfyUI

    Wan Animate 2 lands as an end-to-end character animation framework that drops the intermediate motion extractor. Another model in the weekly rotation, with a genuinely different architecture behind it.

    View source ↗
  5. 11d ago

    Seedance 2.5 is now available via Partner Nodes

    Seedance 2.5 arrives as a Partner Node with 30-second single-shot generation and up to 50 reference assets. Hosted rather than local, consistent with how closed models reach the graph.

    View source ↗
  6. 13d ago

    FLUX 3 is now available via Partner Nodes

    FLUX 3 joins as a Partner Node covering video, audio, image and action from one model. Broadens what a single node can produce without changing the runtime.

    View source ↗