LTX 2.3 Looping Sampler Upscale Optimization Workflow

This workflow is designed for LTX 2.3 looping sampler upscaling and video enhancement optimization. Its main purpose is to take an existing video, extract its visual frames, audio, and frame rate, then process it through an LTX 2.3 refinement pipeline using tiled VAE encoding, looping sampling, conditional guidance, detail LoRAs, large-size image scaling, and final video reconstruction. It is built for creators who want to improve video clarity, detail density, and temporal consistency without rebuilding a clip from scratch.

The workflow uses an LTX 2.3 distilled model route, with ltx-2.3-22b-distilled as the main checkpoint base. It also includes Gemma 3 12B text encoding, LTX video VAE, LTXVLoopingSampler, LTXVSpatioTemporalTiledVAEDecode, VAEEncodeTiled, GetVideoComponents, CreateVideo, SaveVideo, and multiple LoRA enhancement routes. The workflow loads ltx-2-19b-distilled-lora-384 and ltx-2-19b-ic-lora-detailer, which indicates that the graph is focused on refinement, consistency, and detail recovery rather than only basic generation.

A key part of this workflow is the looping sampler. Standard video enhancement often suffers from temporal seams, flickering, or inconsistent detail from one segment to the next. LTXVLoopingSampler is used here to process the video in a more controlled temporal structure. It supports temporal tile size, temporal overlap, guiding strength, conditional image strength, spatial tiling, spatial overlap, and optional guiding latents. This makes the workflow more suitable for longer video refinement, loop-style video polishing, and high-resolution enhancement where memory pressure would otherwise become a problem.

The workflow also keeps the original video structure. GetVideoComponents extracts the source images, audio, and FPS from the input video. The image frames are scaled to a larger target size through ImageScaleToMaxDimension, with the example setting using a largest size of 2560. After that, the frames are encoded through tiled VAE encoding, refined through the LTX looping sampler, decoded through spatio-temporal tiled VAE decoding, and rebuilt into a final video with the original audio and frame rate.

The tiled encode / decode design is important. High-resolution video processing is memory-heavy, especially when the video contains many frames. By splitting the video spatially and temporally, the workflow can improve resolution and detail while keeping the process more stable. This is useful for creators working with AI-generated video previews, cinematic clips, character videos, social media loops, Bilibili demos, YouTube examples, and Civitai workflow showcases.

The workflow also uses manual sigma settings such as 0.85, 0.7250, 0.4219, and 0.0, showing that the sampling route is tuned for controlled refinement rather than aggressive regeneration. The goal is to enhance the source video while reducing unwanted anatomical errors, distorted bodies, broken limbs, awkward poses, and unstable motion through a strong negative prompt.

This workflow is ideal for LTX 2.3 video upscale testing, looping sampler experiments, video detail enhancement, AI short-video polishing, temporal consistency improvement, and final-output optimization before publishing. If you want to see how the video input, LTX 2.3 looping sampler, tiled VAE encoding / decoding, detail LoRA, audio preservation, and final optimized video export are connected, watch the full tutorial from the YouTube link above.

⚙️ Try the Workflow Online

👉 Workflow: https://www.runninghub.ai/post/2041055276513628161?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/BV1gaSfBgEqz/

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/2041055276513628161?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/BV1gaSfBgEqz/

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。