comfyui-extend

ComfyUI/custom_nodes/ComfyUI-LTXFlow

Then restart ComfyUI.

Download Models

From this repo:

COMFYUI_DIR=/workspace/ComfyUI bash scripts/download_models.sh

Change COMFYUI_DIR to your actual ComfyUI path. The script downloads every model needed by:

workflows/05_qwen_edit_to_ltx_first_last.json

It supports Hugging Face auth if needed:

HF_TOKEN=hf_xxxxx COMFYUI_DIR=/workspace/ComfyUI bash scripts/download_models.sh

This script only downloads model files. It does not install ComfyUI custom node packs.

Install Workflow Custom Nodes

The Qwen Edit to LTX workflow also depends on custom node packs outside this repo. If ComfyUI shows an "Installation Required" dialog for workflows/05_qwen_edit_to_ltx_first_last.json, install these packs with ComfyUI Manager:

• ComfyUI-KJNodes - provides LazySwitchKJ, VAELoaderKJ, and LTX2SamplingPreviewOverride
• WhatDreamsCost-ComfyUI - provides LTXSequencer
• ComfyUI Easy Use - provides easy mathInt

Manual install:

COMFYUI_DIR=/workspace/ComfyUI bash scripts/install_custom_nodes.sh

Change COMFYUI_DIR to your actual ComfyUI path. The helper clones or updates:

• https://github.com/kijai/ComfyUI-KJNodes
• https://github.com/WhatDreamsCost/WhatDreamsCost-ComfyUI
• https://github.com/yolain/ComfyUI-Easy-Use

Restart ComfyUI after installing custom nodes, then refresh the browser page.

Frontend

Custom frontend code lives here:

js/ltxflow_ui.js

The package exposes it through:

WEB_DIRECTORY = "./js"

This frontend adds node colors, short in-node notes, and default widget adjustments for LTXFlow nodes. The main node interfaces still come from each Python node's INPUT_TYPES, which is how ComfyUI builds standard node widgets.

Nodes

Proper LTX 2.3 first/last workflow, based on the WhatDreamsCost reference workflow:

workflows/04_ltxflow_first_last_wdc_ltx23.json

Use this one for the first real test. It requires the same dependencies as the reference workflow:

• WhatDreamsCost-ComfyUI
• official LTX/ComfyUI LTX nodes used by that workflow
• VideoHelperSuite SaveVideo/video combine dependencies
• this ComfyUI-LTXFlow package

Qwen Image Edit to LTX first/last workflow:

workflows/05_qwen_edit_to_ltx_first_last.json

This is the main Qwen + LTX workflow. It wires:

source image + reference image
  -> Image Edit (Qwen-Image 2511)
  -> LTX Flow - Qwen Edit Bridge
  -> LTX Flow - First/Last Guide
  -> LTX sampler/video output

The bridge sends the original source image as the LTX first frame and the Qwen-edited image as the LTX last frame.

Required custom node packs:

• ComfyUI-KJNodes for LazySwitchKJ, VAELoaderKJ, and LTX2SamplingPreviewOverride
• WhatDreamsCost-ComfyUI for LTXSequencer
• ComfyUI Easy Use for easy mathInt

Required Qwen models from the official ComfyUI template:

• qwen_2.5_vl_7b_fp8_scaled.safetensors in ComfyUI/models/text_encoders/
• Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors in ComfyUI/models/loras/
• qwen_image_edit_2511_bf16.safetensors in ComfyUI/models/diffusion_models/
• qwen_image_vae.safetensors in ComfyUI/models/vae/

Required LTX 2.3 models:

• ltx-2.3-22b-dev-fp8.safetensors in ComfyUI/models/checkpoints/
• gemma_3_12B_it_fp8_scaled.safetensors in ComfyUI/models/text_encoders/
• ltx-2.3-22b-distilled-lora-384.safetensors in ComfyUI/models/loras/
• ltx-2.3-spatial-upscaler-x2-1.1.safetensors in ComfyUI/models/latent_upscale_models/

You do not

Vérifié 12 sept.