TeleStyle Video Loader

This node is the least exciting thing in the pack, and it's also where the pack's hidden cost lives. It does one job: decide which Wan 2.1 base model you're building on, and make sure every weight it needs is on disk. That second half is a first-run download measured in gigabytes, and it's the reason your graph looks frozen at this node the first time you queue it.

What it actually does

TeleStyleVideoLoader is a path resolver wearing a loader costume. Give it wan_model - one of the two choices, Wan2.1-T2V-1.3B or Wan2.1-T2V-14B - and it downloads three things into ComfyUI/models/TeleStyle/:

• the full Wan 2.1 T2V Diffusers snapshot for your chosen tier,
• dit.ckpt, TeleStyle's style-transfer transformer weights from the Tele-AI/TeleStyle repo,
• prompt_embeds.pth, precomputed text embeddings the pipeline reuses instead of running a T5 encoder.

Then it hands you a model output - a plain dictionary of those paths, nothing loaded into VRAM. All the actual model loading happens lazily inside TeleStyleVideo when you queue the graph. So "loading" here really means "downloading and resolving." Your cue that it worked is the console filling with [TeleStyle] Downloading … lines.

Pick the 1.3B tier to stay under roughly 16 GB VRAM, or the 14B if you've got 40 GB+ to burn and want the quality Wan 2.1 is actually known for. If you're on mid-range hardware, 1.3B is the sane default - it's the low-VRAM tier the community treats as the "test the workflow before committing" option.

Install

ComfyUI Manager can find the pack if you search its title, Comfyui_SynVow_TeleStyle. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/shumoLR/Comfyui_SynVow_TeleStyle
cd Comfyui_SynVow_TeleStyle
pip install -r requirements.txt

That requirements.txt is the real gotcha. It pins modern versions - transformers>=4.57.3, diffusers>=0.36.0, peft>=0.17.0 - plus decord, opencv-python, sentencepiece, einops, omegaconf. Since every custom node installs into the same Python environment, these pins can fight other video packs you already have. If something unrelated stops importing after this install, that's the conflict to suspect first.

One thing to know before you commit: this is a fresh, one-commit port with essentially zero community footprint. The better-known ComfyUI TeleStyle port (ComfyUI-TeleStyle) is the one with the Reddit threads and the 6 GB claim. Don't expect much help in issues if something breaks here.

The first queue will stall at this node for a long time. That's normal - it's pulling the Wan snapshot plus the TeleStyle weights over the network. Leave it alone, and don't mistake it for a hang.