Prompt Relay Story Director + Dual-Character LoRA Format Workflow

This workflow is designed as a Prompt Relay story director for LTX 2.3, with special support for dual-character LoRA prompt formatting. Its main purpose is to take messy user input, images, short ideas, dialogue drafts, storyboard references, or incomplete Chinese / English descriptions, and convert them into a clean final Prompt Relay format using only two fields: global_prompt and local_prompts.

Unlike a normal prompt-writing workflow, this setup is not only generating a beautiful description. It is acting like a structured video director. It analyzes the input image or idea, decides what should remain stable across the whole video, separates global visual anchors from time-based actions, organizes dialogue if needed, and outputs a format that is easier to use in LTX 2.3 Prompt Relay Encode workflows.

The workflow uses an image input route, an LLM API node, a text input node, and an output display node. The uploaded workflow instruction is highly specific: it tells the LLM to read all provided input, including images, multiple images, storyboard grids, rough ideas, Chinese phrases, existing prompts, dialogue drafts, duration requests, no-dialogue requests, and style requests. Then it rewrites everything into a usable LTX 2.3 Prompt Relay structure.

The core rule is simple: global_prompt locks the whole video, while local_prompts controls the timeline. The global_prompt should contain persistent full-video anchors such as scene, characters, identity, left-right placement, costume, style, lighting, atmosphere, camera baseline, dialogue status, audio style, sound design baseline, stability constraints, and negative restrictions. The local_prompts should contain only time-specific events such as camera movement, action, dialogue, reactions, gestures, object movement, environmental motion, transitions, and time-specific sound effects.

A major strength of this workflow is dual-character handling. If the input contains two main characters, the workflow automatically follows LTX 2.3 dual-character LoRA rules: stable identity, fixed left-right positioning, natural eye contact, subtle gestures, controlled body motion, clear interaction logic, no face swapping, no identity drift, no left-right position swapping, and no duplicated characters. This is especially useful because two-character video generation is often unstable when prompts are too loose.

The workflow also includes clear dialogue logic. If the user provides dialogue, the LLM organizes, shortens, humanizes, and completes the lines so they sound natural and suitable for lip sync. If the user does not ask for dialogue, the workflow does not force speech. If the user explicitly requests silence, no dialogue, roaring, breathing, or sound effects only, the workflow generates a no-dialogue version with action and sound design instead.

Timing is also standardized. A 5-second video uses four short segments, a 10-second video uses four 2.5-second segments, and a 20-second video uses five 4-second segments. This makes the output easier to paste into Relay-style workflows and reduces prompt chaos.

The example output in the workflow creates a 10-second dual-character scene in a Chinese teahouse courtyard: a Daoist tea master on the left and a steampunk inventor on the right. The scene includes stable identities, clear left-right roles, Mandarin dialogue, ambient sound design, subtle hand gestures, portal effects, talisman effects, and controlled camera movement. This shows the intended use clearly: the workflow turns rough creative input into a structured cinematic prompt with scene logic, dialogue logic, sound logic, and motion continuity.

This workflow is ideal for AI video creators who want to build better LTX 2.3 Prompt Relay scripts, dual-character dialogue prompts, story-driven short clips, AI drama scenes, character interaction tests, storyboard-to-prompt conversion, and reusable prompt-director templates. If you want to see how image understanding, dialogue organization, Prompt Relay timing, and dual-character LoRA stability rules work together, watch the full tutorial from the YouTube link above.

⚙️ Try the Workflow Online

👉 Workflow: https://www.runninghub.ai/post/2048717560673214466?inviteCode=rh-v1111

Open the link above to run the workflow directly online and view the generation results in real time.

If the results meet your expectations, you can also deploy it locally for further customization.

🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!

📺 Bilibili Updates (Mainland China & Asia-Pacific)

If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.

📺 Bilibili Video: https://www.bilibili.com/video/BV1za5y6FE7r/

I will continue updating model resources on Quark Drive:

👉 https://pan.quark.cn/s/20c6f6f8d87b

These resources are mainly prepared for local users, making creation and learning more convenient.

⚙️ 在线体验工作流

👉 工作流： https://www.runninghub.ai/post/2048717560673214466?inviteCode=rh-v1111

打开上方链接即可直接运行该工作流，实时查看生成效果。

如果觉得效果理想，你也可以在本地进行自定义部署。

🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！

📺 Bilibili 更新（中国大陆及南亚太地区）

如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。

📺 B站视频： https://www.bilibili.com/video/BV1za5y6FE7r/

我会在 夸克网盘 持续更新模型资源：

👉 https://pan.quark.cn/s/20c6f6f8d87b

这些资源主要面向本地用户，方便进行创作与学习。