LTX 2.3 Three-Stage Multi-Image First/Last Frame Workflow

This workflow is designed for LTX 2.3 multi-image first-frame / last-frame video generation with a three-stage sampling structure. Its main purpose is to connect multiple reference images into a more coherent video sequence, while using staged LTX 2.3 refinement to improve motion stability, visual continuity, texture quality, and final frame consistency.

Unlike a simple image-to-video workflow that only uses one starting image, this setup is built around stronger keyframe guidance. The workflow can use reference images as visual anchors so the video does not drift too far away from the intended subject, composition, or final state. This makes it useful for first-frame / last-frame animation, multi-image transition shots, character transformation clips, scene evolution, cinematic loops, and AI short-video production where the creator needs more control than a single-image prompt can provide.

The workflow uses ltx-2.3-22b-dev as the main video model, Gemma 3 12B text encoding, LTX video VAE, LTX audio VAE, LTXVConditioning, EmptyLTXVLatentVideo, LTXVEmptyLatentAudio, LTXVConcatAVLatent, LTXVSeparateAVLatent, SamplerCustomAdvanced, ManualSigmas, LTXVLatentUpsampler, tiled VAE decoding, and preview / output nodes. It also loads the ltx-2.3-22b-distilled-lora-384 route, which helps strengthen the LTX 2.3 generation process for this staged workflow.

The strongest part of this workflow is the three-stage sampling design. The first stage builds the main motion and temporal direction from the image references and text prompt. The next stages continue from the generated latent result, using lower sigma ranges such as 0.85, 0.7250, 0.4219, and 0.0 to refine the motion, texture, and continuity rather than completely regenerate the video. This is important because multi-image video generation needs balance: too much freedom may destroy the reference images, while too little freedom may produce weak motion.

The workflow also includes latent upscaling through LTXVLatentUpsampler. Instead of only enlarging the final decoded frames, it improves the video in latent space before final decoding. This helps the output retain better structure, cleaner detail, stronger subject clarity, and more stable transitions across frames. For LTX 2.3 video generation, this is especially useful when the goal is not just to create motion, but to produce a more polished result for publishing.

The negative prompt is focused on suppressing common AI video failures such as low resolution, blur, static frames, no movement, subtitles, overlays, watermarks, scene cuts, scene transitions, warping, extra hands, extra limbs, and unstable body parts. These restrictions make the workflow more suitable for controlled production instead of random animation.

This workflow is ideal for creators who want to push LTX 2.3 beyond a basic one-pass I2V setup. It can be used for multi-image keyframe animation, first-frame / last-frame video generation, smooth transition tests, cinematic AI clips, character transformation videos, RunningHub demos, Civitai previews, and YouTube / Bilibili workflow showcases. If you want to see how the reference images, LTX 2.3 model route, three-stage sampler passes, latent upscaling, and final video output work together, watch the full tutorial from the YouTube link above.

⚙️ Try the Workflow Online

👉 Workflow: https://www.runninghub.ai/post/2033913591946027010?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/BV1WvwkzjERN/

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/2033913591946027010?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/BV1WvwkzjERN/

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。