TV → Hunyuan Adapter 🎬

Let's be straight about Hunyuan Video in 2026: it's a distant third. Wan 2.2 owns the quality crown, LTX owns the speed niche, and Hunyuan's photorealistic lead and its famous 201-frame natural looping have both been eaten by VACE and the Wan side of the ecosystem. If you're not already on Hunyuan, this adapter probably isn't what pulls you over. But if you are - or you're on Hunyuan 1.5 for its step-distilled speed and prompt adherence - TV → Hunyuan Adapter 🎬 gives the Director timeline a first-class Hunyuan path instead of leaving you out in the cold.

Mechanically, Hunyuan conditions like Wan, not like LTX: per-clip. TextEncodeHunyuanVideo_ImageToVideo bundles text plus one reference image into a conditioning object, and EmptyHunyuanLatentVideo sizes a latent for a given frame count. There's no multi-shot guide injection, so the adapter does exactly what the Wan one does - it walks the timeline and emits one conditioning/latent bundle per shot, reusing the same TV_WAN_SHOTLIST shape so the pack's shared iterator works on both.

Inputs and outputs that matter

Required: timeline (the DIRECTOR_TIMELINE from TV Director), clip, and vae. clip_vision is optional, for image-aware conditioning on image-to-video shots.

Outputs: hunyuan_shots (a TV_WAN_SHOTLIST) and shot_count (INT). Same loop pattern as Wan: pull one shot at a time with TV Wan Shot Iterator, sample, then concatenate your clips downstream.

How a shot becomes conditioning

For a shot with a reference image, the adapter calls TextEncodeHunyuanVideo_ImageToVideo with image_interleave=2 - that's how often the image conditioning re-asserts itself through the sequence; 1 would mean every frame, higher is sparser, and 2 is a decent middle ground that favors prompt adherence while still respecting the source image. The reference frame gets VAE-encoded and padded along the temporal axis to the shot's frame count. No image? Then it's a plain CLIPTextEncode plus an EmptyHunyuanLatentVideo sized from the timeline's width/height.

Install and the trap

The pack installs normally:

cd ComfyUI/custom_nodes/
git clone https://github.com/TensorVizion/ComfyUI-Director-Node-Pack tensorvizion-director

Restart. It targets the classic HunyuanVideo node set that ships in ComfyUI core - TextEncodeHunyuanVideo_ImageToVideo + EmptyHunyuanLatentVideo. The trap: Hunyuan 1.5 renames those nodes (think HunyuanVideo15ImageToVideo), so if you're on 1.5 this adapter won't work and will tell you so clearly. It's classic-HunyuanVideo-only, and 1.5 would need its own variant. One more thing worth knowing before you commit: Tencent's Hunyuan license has a territorial clause that touches the EU, UK, and South Korea - output included - so check that before you build a commercial pipeline on it.