RunningHub Workflow

Try the workflow online right now — no installation required.
Workflow: https://www.runninghub.ai/zh-cn/post/2073620739328143361?inviteCode=rh-v1532

If the results is good then you can deploy it locally

🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance 24/48GB(Plus mode)

LTX 2.3 Director 2.0 UHD All-in-one Workflow NSFW

LTX 2.3 WDC Director V5 | Setup Quick Reference
Model file structure
ComfyUI/models/
├─ diffusion_models/ltx/
│ └─ ltx-2.3-22b-distilled-1.1_transformer_only_mxfp8_block32.safetensors
├─ clip/
│ ├─ gemma_3_12B_it_fp8_e4m3fn.safetensors
│ └─ ltx-2.3_text_projection_bf16.safetensors
├─ vae/
│ ├─ LTX23_video_vae_bf16.safetensors
│ └─ LTX23_audio_vae_bf16.safetensors
├─ latent_upscale_models/
│ └─ ltx-2.3-spatial-upscaler-x1.5-1.0.safetensors
└─ loras/LTX/
└─ LTX 2.3 LoRAs selected by the workflow (optional/replaceable)
Important: place the Gemma text encoder under models/clip/, not models/text_encoders/.
Required custom nodes
WhatDreamsCost-ComfyUI — LTX Director timeline and Director Guide
https://github.com/WhatDreamsCost/WhatDreamsCost-ComfyUI

ComfyUI-LTXVideo — LTX AV latents, conditioning, and UHD upscale nodes
https://github.com/Lightricks/ComfyUI-LTXVideo

ComfyUI-KJNodes — Lazy Switch, VAE Loader, sampling preview, and utilities
https://github.com/kijai/ComfyUI-KJNodes

rgthree-comfy — Power LoRA Loader, Set/Get nodes, and group bypass controls
https://github.com/rgthree/rgthree-comfy

ComfyUI-VideoHelperSuite — video encoding, audio muxing, and MP4 output
https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite

ComfyUI-TT-Resolution_selector-Node — provides LHResolutionSetting
This is a locally bundled plugin. Its package metadata contains an invalid placeholder GitHub URL. When moving the workflow, copy the complete plugin folder or replace this node with integer width/height nodes.

Quick notes
If nodes appear red, install/update the required packages, restart ComfyUI, and reload the original JSON.
Saved defaults: 30 fps, 300 frames (10 seconds), 9-step schedule, LCM, CFG 1.0, randomized seed.
Only the first stage is enabled by default; Sampler - 2nd Upscale 1.5x is bypassed.
F-Generate Audio / T-Uploaded Audio: false selects model-generated audio; true selects uploaded audio.
The example image and audio reference the original author's input/whatdreamscost/ files. Replace them with your own assets.
Test motion at a lower resolution first. Fix a good seed before enabling the 1.5× UHD stage.

作品

评论