LTX 2.3 Multi-Image Reference | IC Edit Clean No-Subtitle Workflow

This workflow is designed for LTX 2.3 multi-image reference video generation with IC Edit-style clean-output control, focused on creating stable, subtitle-free AI videos from multiple visual references. Its main purpose is to let creators use several reference images as stronger visual anchors, guide the video generation process with clearer character, scene, object, or keyframe information, and reduce unwanted subtitles, fake captions, random text, watermark-like artifacts, overlays, and noisy UI-style elements in the final output.

Compared with a single image-to-video workflow, this setup gives creators more control over visual consistency. A single image can only define the starting appearance, but multi-image reference can provide more information about the subject, style, pose, scene direction, key visual state, or transition target. This is especially useful when the creator wants to preserve a character identity, maintain a consistent costume, reference multiple visual elements, or guide the video toward a more specific cinematic result instead of letting the model freely drift.

The workflow uses LTX 2.3 as the core video generation backbone, with LTXVConditioning, EmptyLTXVLatentVideo, LTXVEmptyLatentAudio, LTXVConcatAVLatent, LTXVSeparateAVLatent, SamplerCustomAdvanced, ManualSigmas, CFGGuider, RandomNoise, LTX2_NAG, LTXVLatentUpsampler, VAEDecodeTiled, LTXVAudioVAEDecode, CreateVideo, VRAMReserver, PurgeVRAM logic, and image-sequence handling. It also includes staged sampling routes, preview routes, tiled decoding, audio latent handling, and frame extraction logic, making it more like a production workflow than a simple one-pass generation graph.

The main strength of this workflow is multi-reference control. The additional images help the model understand the intended visual direction more clearly: what the character should look like, how the scene should feel, what style should remain stable, and what kind of final video the creator wants. This is useful for AI short videos, fantasy scenes, character motion, product shots, stylized video clips, reference-driven transitions, and cinematic image animation.

The IC Edit-style clean-output direction is another important selling point. Many AI video generations may accidentally produce fake subtitles, random letters, caption bars, logo-like symbols, watermark shapes, or interface-like marks, especially when prompts describe dialogue, cinematic framing, social media scenes, or character performance. This workflow is designed around clean visual output, using negative restrictions and generation control to make the final clip easier to publish directly.

The workflow also uses a staged refinement structure. The first stage builds the base motion and visual direction, while later stages refine the generated latent result with lower sigma ranges. This helps improve motion stability, subject consistency, texture quality, and final coherence without completely destroying the original reference information. Tiled VAE decoding and VRAM cleanup make the workflow more practical for heavier video generation tasks.

This workflow is ideal for creators who want to produce multi-image reference videos, high-consistency character clips, clean no-subtitle AI videos, cinematic short scenes, RunningHub demos, Civitai previews, YouTube examples, and Bilibili workflow showcases. If you want to see how LTX 2.3, multi-image reference control, IC Edit-style clean output, NAG enhancement, staged sampling, tiled decoding, and subtitle-free generation work together, watch the full tutorial from the YouTube link above.

⚙️ Try the Workflow Online

👉 Workflow: https://www.runninghub.ai/post/2054535482951913473?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/2054535482951913473?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。