Enviral Load Text Encoder

Text encoders are the forgotten siblings of the model-loading world. The diffusion model gets all the attention, but every modern stack leans on a standalone CLIP or LLM encoder - Qwen2.5-VL for Qwen-Image and Flux2, the WAN encoder for WAN video, plus all the SD/SDXL/FLUX1 CLIPs - and they all pile into models/text_encoders. Enviral Load Text Encoder (class EnviralLoadTextEncoderFiltered) is the filtered loader from the Enviral Design Node Pack that keeps that pile navigable, while keeping every native option the stock loader gives you.

How it works

Same shared filtering base as Enviral Load Checkpoint and Enviral Load Diffusion Model in this pack: a folder prefix filter, an optional text allow_list, and a dropdown that only shows what passes both. The loading is ComfyUI's standard text-encoder load, driven by the full relative path, with validation up front so a bad selection errors out before it runs.

What makes this one different from the other two filtered loaders is that it keeps the native encoder type enum and optional device choice - no functionality was stripped to add filtering.

The inputs and outputs that matter

• folder - default All Text Encoders, or any folder prefix you have. Instant grouping for qwen/, wan/, clip/.
• clip_name - the encoder dropdown. Also accepts a linked STRING if it matches a known ComfyUI path.
• type - this is the one that matters most. It's the full native list: stable_diffusion, stable_cascade, sd3, stable_audio, mochi, ltxv, pixart, cosmos, lumina2, wan, hidream, chroma, ace, omnigen2, qwen_image, hunyuan_image, flux2, ovis, longcat_image, cogvideox, lens, pixeldit, ideogram4, boogu, krea2, joyimage, mage, minimax. The type tells ComfyUI what architecture it's loading, so get it right: qwen_image for Qwen2.5-VL, wan for WAN, flux2 for Flux2.
• device - default or cpu. Force CPU loading/offloading if you're juggling VRAM.
• allow_list - optional multiline allow-list, same behavior as the other filtered loaders.

Outputs are clip (the loaded text encoder, wired into your CLIP Text Encode / TextEncode node) and text_encoder_list, a STRING of visible options, one per line.

Installing it

In the Enviral Design Node Pack. ComfyUI Manager → search "Enviral Design Node Pack" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/EnviralDesign/comfyUI-enviral-design-node-pack

then restart. Nothing to download for the node itself - it reads models/text_encoders. The pack's only hard Python dependency is color-matcher, used by a color-match node elsewhere in the pack, so you're not signing up for a dependency pile to get a filtered loader.

Where people get burned

• Wrong type = subtle garbage. If the encoder loads but your embeddings come out wrong, check type first. It's the single most common mistake with this node family, and the failure is often silent.
• New encoders don't show up. Same ComfyUI caching as always - press R on the canvas or Refresh Node Definitions, then re-pick the folder.
• A linked STRING that doesn't match. Validation fails with a clear message. It has to match a path ComfyUI actually knows.

This is one of those "you don't need it until you need it" nodes - but once models/text_encoders gets past a dozen files, a filtered encoder loader starts feeling like a quality-of-life upgrade rather than a nicety.