DaSiWa V4 and Director Node 2.0 on LTX 2.3: One Timeline for Prompts, Images, Video, Audio and IC-LoRA
DaSiWa V4 and Director Node 2.0 on LTX 2.3: One Timeline for Prompts, Images, Video, Audio and IC-LoRA This episode looks at another LTX 2.3 branch: the DaSiWa (Master Frog) V4 unlocked specialty model, paired with the
DaSiWa V4 and Director Node 2.0 on LTX 2.3: One Timeline for Prompts, Images, Video, Audio and IC-LoRA
This episode looks at another LTX 2.3 branch: the DaSiWa (Master Frog) V4 unlocked specialty model, paired with the author's updated OmniverseForge integrated workflow. The notes below come from the episode subtitles and the published mirror page; no workflow JSON was shipped or parsed, so no node-level parameters are claimed beyond what the episode states.
What DaSiWa V4 is
- An unlocked specialty model for LTX 2.3, updated to V4.
- The author says it supports unrestricted generation.
- The author also updated the OmniverseForge integrated workflow.
- The mirror bundles the model and the workflow.
The Director node 2.0
The workflow mainly relies on the Director node 2.0, an aggregating condition-input node. You feed inputs on the main track, audio track and IC-LoRA track, then set the timeline's start and end times (to cut a segment; can start at 0), plus a frame rate (default 24 fps) and a size in the lower left measured in megapixels.
The concept: the Director node unifies prompts, images, video, audio and IC-LoRA control onto one timeline. It has no single fixed mode โ
- Main track empty โ text-to-video.
- One image โ image-to-video.
- Last image set as the tail frame โ first/last frame mode.
- Multiple images โ multi-keyframe.
- A normal video โ video repaint or continuation.
- Audio track โ audio input; with Inpaint on, existing audio is kept and blank regions are filled in; with it off, the behaviour differs.
Hardware and deployment
- LTX 2.3 has been optimized by the author: 24 GB can run it.
- The author recommends a 48 GB 4090.
- The mirror is on ไผไบๆบ็ฎ: https://www.compshare.cn/images/iu8Whe0MA5De?referral_code=A7fTQUYY9lCDFJwA6MMvoq
- Open the mirror page.
- Deploy an instance and wait for the image to come up.
- Open ComfyUI, pick the workflow from the left panel, and use the Director node.
Links
- Mirror (with referral code): https://www.compshare.cn/images/iu8Whe0MA5De?referral_code=A7fTQUYY9lCDFJwA6MMvoq
- Video tutorial (YouTube): https://www.youtube.com/watch?v=5mE6Wkx3NHg
- Chinese original (Bilibili): https://www.bilibili.com/video/BV1JNgx6wEgj
Limits
- "Unrestricted generation" is the author's description of the unlocked model; usage still depends on the platform and the model's own terms.
- Timeline parameters (24 fps default, megapixel sizing) are the workflow's defaults, not universal settings.
- The mirror is a deployment convenience; third-party nodes and LoRAs keep their own licenses.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes โ full credit and traffic to the original publisher.