Skip to content

Models

What each part of Castcut needs in ComfyUI. Install them from Settings → ComfyUI → Models (filter by the tool); Settings → Overview → Play checks says what is missing.

Start small

For a first film you need one still engine and LTX-2.5 for clips. Add the rest when you want the tool.

Stills — pick at least one

Engine Size Good for
Qwen Rapid AIO (SFW or NSFW v23) 28 GB Fastest stills; Intimate / Raunchy need the NSFW one
Qwen-Image-Edit 2511 + Lightning 41 GB + 0.8 GB The liveliest clothed stills; room photos (also on Rapid); shoe pictures
Qwen-Image 2.1 + Turbo 8-step (or Pruna 8-step) — the default when installed 14 GB + 0.9 GB Closest likeness. Turbo is the Qwen team's own fast version (Qwen Research License) — sharper, and clean lying down where Pruna can melt

Clips

What Models Size
Clips, talking clips, Add voice LTX-2.5 (5 files) — accept the terms on Hugging Face and set HF_TOKEN ~40 GB
Her kept voice LTX-2.3 ID-LoRA 1.2 GB
Two-person adult clips WAN Rapid AIO NSFW 23 GB
MiniMax-H3 (optional, best motion) 5 files — licence excludes the US, EU, UK and South Korea ~45 GB

Film

What Models Size
Score this film ACE-Step 1.5 10 GB
Room sound under talking castcut-nodes 1.7.0+ (a node pack) —
Face finish ComfyUI-ReActor + CodeFormer small

Finishing tools

Tool Models Size
Sharpen & smooth · Upscale to 4K SeedVR2 3B int8 + VAE (RIFE 4.26 for clips) 4 GB
Copy moves Wan Animate 2 distilled int8, CLIP vision H, Wan 2.1 VAE, UMT5 25 GB
Dance to music Wan-Dancer global + local, lightx2v LoRA, CLIP vision H, Wan 2.1 VAE, UMT5 46 GB
Lip-sync Wan 2.1 I2V 480p fp8, InfiniteTalk (single; multi for two people), lightx2v LoRA, wav2vec2, UMT5 29–34 GB
Relight Qwen-Edit Relight LoRA (+ a Rapid AIO checkpoint) 0.2 GB
Sketch A Rapid AIO checkpoint —

Shared files (UMT5, CLIP vision, the Wan 2.1 VAE, lightx2v) download once.

Checks

Check Needs
Pose check, pose from a photo comfyui_controlnet_aux (DWPose)
Face match, Face finish ComfyUI_FaceAnalysis
Outfit, place, adult-age and Auto-review checks A vision LLM — see Install

Heal & ready installs the node packs for you.