Nodes/h4_Live/đź”— h4_Link (Civitai Bridge)
ComfyUI Node

đź”— h4_Link (Civitai Bridge)

Civitai, but you never leave the canvas

By m3rr·Created 9 months ago·Updated 10 days ago· 1
đź”— h4_Link (Civitai Bridge)
    • model_name
    • model_path
    • trigger_words
    • civitai_url
    • json_metadata
    â—„active_model_nameâ–ş
    â—„inject_modeAuto (LoRA/Checkpoint)â–ş
    â—„model_typeLoRAâ–ş
    â—„trigger_wordsâ–ş

    Every model hunt goes the same way: you find something good on Civitai, download it into your Downloads folder, realize it went to the wrong place, manually move it into models/loras, and then hand-type the filename into a loader. The h4_Link (Civitai Bridge) node exists to kill that whole ritual. Search Civitai from inside ComfyUI, preview the showcase images, download straight into the correct model folder, and push the filename directly into your loader node. The browser never opens.

    It's the newest flagship of the h4_Live pack and the one with the most moving parts. This is also the rare node where the "node" itself is almost a trophy - the real action lives in the frontend drawer and a couple of server routes.

    How it works

    The backend talks to Civitai's public REST API (civitai.com/api/v1) over its own aiohttp routes. Search queries filter by model type (LoRA, Checkpoint, VAE, ControlNet, Embeddings, UNet) and sort by rating or downloads. Downloads stream in the background in 1MB chunks so your UI doesn't freeze, and ComfyUI's own folder-path API decides the destination - LoRAs land in models/loras, checkpoints in models/checkpoints, and so on.

    Two nice touches: it writes a .txt sidecar next to the model with the trained trigger words, plus a .json with the raw Civitai version payload - so the metadata survives even if you never open the site again. And the "Load into Node" button walks your selected loader's widgets and drops the filename in for you.

    The inputs and outputs

    The node itself is mostly a mirror of what the drawer is doing. active_model_name shows the filename you selected or injected; inject_mode lets you restrict injection to LoRA loaders, checkpoint loaders, or auto-detect. The model_type enum and trigger_words are optional and self-explanatory.

    It outputs five strings, which is useful if you want the rest of your graph to know what you're running: model_name, model_path, trigger_words, civitai_url, and json_metadata. You can wire these into text display, filename builders, or note nodes. It's an output node, so it also serves as a graph endpoint.

    Installing it

    Same as every h4_Live node - install the pack:

    cd ComfyUI/custom_nodes
    git clone https://github.com/m3rr/h4_Live
    # restart ComfyUI
    

    or ComfyUI Manager → search "h4_Live." No API key needed; the name "bridge" is a lie in the best way - it's a read-only client of Civitai's public API.

    Where it can bite you

    Civitai rate-limits hard. The README is upfront that you'll hit HTTP 429s and timeouts on search, thumbnails, and big downloads - that's Civitai's infrastructure throttling you, not a bug in the pack. If a download stalls mid-way, check the status route before re-downloading. And the frontend drawer is exactly the kind of "tactical flair" that the pack's own README warns may be degraded in ComfyUI Desktop Edition's sandboxed rendering - if the drawer won't open, that's the first suspect.

    Worth saying: this is a young, single-author pack with no community footprint to speak of yet, and the bridge is its newest module. It's genuinely useful for the "download one LoRA while building a workflow" flow - just don't expect it to replace a serious model manager overnight.

    Categoryh4_Live/QoL

    Inputs (4)

    NameTypeDefaultDescription
    active_model_nameSTRINGThe filename of the model selected or injected from Civitai.
    inject_modeCOMBOAuto (LoRA/Checkpoint)3 options: Auto (LoRA/Checkpoint), LoRA Loader Only, Checkpoint Loader Only
    model_typeoptCOMBOLoRA6 options: LoRA, Checkpoint, VAE, ControlNet, Embeddings, UNet
    trigger_wordsoptSTRINGAuto-extracted trigger words for the downloaded model.

    Outputs (5)

    NameTypeDescription
    model_nameSTRING—
    model_pathSTRING—
    trigger_wordsSTRING—
    civitai_urlSTRING—
    json_metadataSTRING—