Watch the full video first if you want to understand how this LTX2.3 3.5 one-click controlled video generation workflow works in practice. The video shows how a creator can use one unified control pipeline to generate a video through LTX2.3, guided conditioning, staged latent processing, and final three-stage rendering.
This ComfyUI workflow is designed as a one-click video generation system for LTX2.3 3.5. Its purpose is not only to generate a basic text-to-video result, but to give creators a more controlled production route where input preparation, prompt conditioning, video latent construction, staged sampling, latent upscaling, decoding, audio handling, and final export are arranged into one complete workflow.
The workflow is built around ltx-2.3-22b-dev_transformer_only_fp8_scaled.safetensors as the main LTX2.3 video model. The text encoding route uses gemma_3_12B_it_fp8_e4m3fn.safetensors together with ltx-2.3_text_projection_bf16.safetensors through a DualCLIPLoader. The video VAE uses LTX23_video_vae_bf16.safetensors, while the audio route uses LTX23_audio_vae_bf16.safetensors. This gives the workflow a more modular LTX2.3 architecture than a simple checkpoint-only graph.
The input preparation section uses image resizing, video information reading, LTXVPreprocess, and size-matching logic to prepare the source material before generation. The workflow can read video information such as FPS and frame count, then build matching latent structures through EmptyLTXVLatentVideo and LTXVEmptyLatentAudio. This makes the workflow better suited for controlled video production where frame count, frame rate, image size, and final output behavior need to stay predictable.
The control section uses LTXAddVideoICLoRAGuide to inject image/video guide information into the LTX generation process. This allows the model to follow the intended visual direction more closely than pure text prompting. The workflow also includes optional control modules such as DepthCrafter, CannyEdgePreprocessor, and DWPreprocessor. These are prepared as optional or bypassed control routes, giving the workflow room to expand into depth control, edge control, or pose control when needed.
The generation route is arranged around a staged rendering structure. The first stage builds the base video structure and motion. The second stage separates audio and video latents, refines the video latent, and prepares it for higher-quality continuation. The third stage uses euler_cfg_pp sampling with a compact sigma schedule, then decodes the final result through VAEDecodeTiled for memory-friendly output.
The audio-video structure is handled through LTXVSeparateAVLatent and LTXVConcatAVLatent. This lets the workflow split video and audio latent components, process them separately, and recombine them before later stages. LTXVAudioVAEDecode is used to decode the audio route, and VHS_VideoCombine exports the final video in a RunningHub-safe H264 MP4 format.
This workflow is suitable for creators who want one main LTX2.3 video generation app instead of several fragmented test graphs. It can be used for cinematic short clips, image-guided video, prompt-driven video, controlled motion experiments, AI video thumbnails, short-form creative content, and staged HD video generation.
Main features:
LTX2.3 3.5 one-click video generation workflow
Unified controlled video generation route
Three-stage rendering structure
ltx-2.3-22b-dev_transformer_only_fp8_scaled.safetensors support
Gemma3 FP8 text encoder support
LTX text projection support
LTX23 video VAE support
LTX23 audio VAE support
LTXVPreprocess input preparation
LTXAddVideoICLoRAGuide guide injection
LTXVConditioning video-aware prompt route
EmptyLTXVLatentVideo target latent creation
LTXVEmptyLatentAudio audio latent creation
LTXVSeparateAVLatent audio-video split
LTXVConcatAVLatent audio-video reconstruction
LTXVLatentUpsampler refinement route
SamplerCustomAdvanced staged sampling
euler_cfg_pp final refinement
VAEDecodeTiled memory-friendly decoding
LTXVAudioVAEDecode audio decoding
Optional DepthCrafter control route
Optional Canny edge control route
Optional DW Pose control route
VHS_VideoCombine final MP4 export
Suggested workflow:
Start with a clear prompt and a clean visual reference if you are using image or video guidance. Keep the creative direction focused: subject, action, camera movement, lighting, atmosphere, and visual style should not conflict with each other. Use the first stage to confirm that the basic motion and scene direction are correct. Use the later stages for cleaner detail, stronger continuity, and more polished output. If the video becomes unstable, reduce aggressive motion language and simplify the prompt. If you need stronger structure control, enable only one additional control route at a time, such as depth, edge, or pose, instead of stacking too many controls at once.
⚙️ RunningHub Workflow
Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2068580767306964993?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/2068580767306964993?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。