Watch the full video first if you want to understand how this LTX2.3 3.5 storyboard director workflow controls character, action, and scene continuity. The video shows how LTXDirector can be used as a more structured director system, helping the generated video follow planned character behavior, camera logic, and scene progression instead of drifting randomly.
This ComfyUI workflow is designed for LTX2.3 3.5 character-action-scene video generation with storyboard director control. Its main purpose is to keep a video sequence coherent across multiple shots. In normal text-to-video generation, the character may change appearance, the action may drift, or the background may become inconsistent. This workflow is built to reduce that problem by combining LTXDirector, LTXDirectorGuide, negative guidance, staged rendering, and high-definition latent refinement.
The workflow is built around ltx-2.3-22b-dev-dare-ties-distilled-1.1.safetensors as the main LTX2.3 checkpoint. The text encoding route uses gemma_3_12B_it_fp8_e4m3fn.safetensors through the LTX AV text encoder loader. The workflow also uses LTXVAudioVAE, LTXVConditioning, LTX2_NAG, LTXDirector, LTXDirectorGuide, SamplerCustomAdvanced, LTXVLatentUpsampler, VAEDecodeTiled, LTXVAudioVAEDecode, and final video output nodes.
The director section is the core of this workflow. LTXDirector organizes the video as a directed timeline rather than a single uncontrolled generation pass. This makes it possible to describe the character, action, scene, framing, and timing in a more structured way. The goal is to keep the same character from scattering across shots, keep the action readable, and keep the scene direction consistent from beginning to end.
The workflow uses LTXDirectorGuide in multiple stages. In Stage 1, the director guide helps establish the base composition, character position, shot rhythm, and main motion. In Stage 2, the video latent is refined after spatial upscaling while preserving the director plan. In Stage 3, another guided refinement pass is used for final visual stability and high-definition polish.
The negative prompt route is built to suppress common video problems such as low quality, blurry output, temporal flicker, frame jitter, ghosting, warped geometry, sudden scene cuts, identity drift, face distortion, bad hands, inconsistent clothing, extra people, subtitles, watermarks, readable letters, speech bubbles, cluttered details, and unwanted prop chaos. This makes the workflow especially useful for character-focused video production.
The model chain also includes a director-oriented LoRA rack, with slots for cinematic style, identity or reward alignment, visual-temporal stability, motion-track control, watermark or subtitle cleanup branches, and final crispness enhancement. This gives the workflow a more production-oriented structure than a simple LTX2.3 test graph.
The rendering route uses a three-stage structure. Stage 1 creates the initial director-guided video. Stage 2 uses the LTX2.3 spatial upscaler x2 1.1 for latent-level refinement. Stage 3 performs the final light sampling pass before tiled VAE decoding. The audio-video latent route is handled through LTXVSeparateAVLatent and LTXVConcatAVLatent, keeping video and audio components organized during refinement.
Main features:
LTX2.3 3.5 storyboard director workflow
Character, action, and scene control
LTXDirector timeline-based video direction
LTXDirectorGuide guided rendering
Independent guide control across Stage 1, Stage 2, and Stage 3
ltx-2.3-22b-dev-dare-ties-distilled-1.1 support
Gemma3 FP8 text encoder support
LTXVAudioVAE support
LTXVConditioning video-aware prompt route
LTX2_NAG quality and stability constraint
Three-stage rendering pipeline
Stage 1 director-guided base generation
Stage 2 latent upscale refinement
Stage 3 high-definition final polish
LTX2.3 spatial upscaler x2 1.1 support
Audio-video latent separation and recombination
VAEDecodeTiled memory-friendly decoding
Designed to reduce identity drift and shot drift
Suitable for character videos, short films, and controlled scene generation
Suggested workflow:
Prepare the character concept, action plan, and scene direction before running the workflow. Write prompts that clearly separate who the character is, what the character is doing, where the scene takes place, and how the camera should move. Avoid adding too many conflicting actions in one segment. Use the first stage to check whether the character, action, and scene direction are correct. If the character identity drifts, strengthen the identity preservation wording and simplify the motion. If the action becomes unclear, make the action chain shorter and more direct. Use the second and third stages only after the base direction is already correct.
⚙️ RunningHub Workflow
Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2068718478793728001?inviteCode=rh-v1111
If the results meet your expectations, you can later deploy it locally for customization.
🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!
📺 Bilibili Updates (Mainland China & Asia-Pacific)
If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.
📺 Bilibili Video: https://www.bilibili.com/video/BV1xw7F6XE7K/
☕ Support Me on Ko-fi
If you find my content helpful and want to support future creations, you can buy me a coffee ☕.
Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.
👉 Ko-fi: https://ko-fi.com/aiksk
💼 Business Contact
For collaboration or inquiries, please contact aiksk95 on WeChat.
⚙️打开下方链接即可在线体验，无需安装。
👉 工作流： https://www.runninghub.ai/post/2068718478793728001?inviteCode=rh-v1111
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/BV1xw7F6XE7K/
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。