Ino Get Clip Download Config

Text encoders are the most forgettable part of a modern model setup - and the easiest to get wrong. You need the right CLIP file from the right repo, in the right subfolder, and a typo in any of it means a failed download or a silent mismatch. Ino Get Clip Download Config removes the typing entirely: you pick a CLIP model from a dropdown, and it hands you a complete download configuration as a JSON string, ready to feed the pack's download nodes.

It's part of the Ino Nodes pack from nobandegani, and it fits into the pack's "config-driven" model management story: build a config once, let download and load nodes consume it. This node is one of a family - VAE, ControlNet, LoRA and image/video model variants all work the same way, reading from the same CSV files.

How it works

The dropdown isn't scraped live. The pack ships a set of CSV files in its data directory - here clip_files.csv - and each row is one known-good CLIP model with its repo_id, filename, subfolder, weight_type (fp16, fp8, bf16...) and target model_type. The node reads that CSV, lists the model names as your dropdown, and when you pick one it serializes the whole row to JSON and hands it out of the single output, ModelConfig.

What's in the config matters because of what consumes it. Wire ModelConfig into Ino Hugging Face Download Model and it knows exactly where to fetch the file and where to put it in ComfyUI/models/text_encoders/. Feed it to Ino Handle Download And Load Model and the whole download-then-load chain is driven by the string alone. The CSVs ship with the pack, and the entries are real - things like the Flux text encoders (clip_l.safetensors, t5xxl_fp16.safetensors) and Wan UMT5 variants - so you're picking from a curated list rather than remembering filenames.

Only two inputs: enabled (the pack's standard toggle) and model (the dropdown). No tokens, no URLs, no API keys.

Where it fits

If you're assembling a Flux or Wan workflow the "Ino way," you collect a ModelConfig string for each piece - CLIP here, VAE from the sibling node, the diffusion model from its own config - then hand them to the download/load machinery. The win is reproducibility: the config string carries every parameter the downloader needs, so a workflow can be shared or re-run without anyone guessing which file goes where.

The honest note: this only covers models the pack's CSV knows about. New or niche text encoders won't be in the dropdown, and you'd go straight to the download node with manual fields instead. For the mainstream encoders, though, it's the no-typo path.

Installing it

It ships with the whole Ino Nodes pack. ComfyUI Manager: search "ComfyUI Ino Nodes", install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/nobandegani/ComfyUI-InoNodes.git
cd comfyui_ino_nodes
pip install -r requirements.txt

Restart after. The usual pack caveats: it's built on ComfyUI's V3 schema (README wants v0.18.1+), and requirements.txt installs inopyutils, the author's helper library that every node imports.

Common issues

If the dropdown looks empty or short, it's populated from the pack's CSV at load time - a bad install or an updated pack that hasn't reloaded can leave it stale, so restart ComfyUI after updating. And remember the output is a JSON string: don't paste it into a field expecting just a filename; it's meant for the pack's own config-consuming nodes. Search "Ino Get Clip Download Config" in the node menu (grouped under InoModelHelper).