Watch the full video first if you want to understand how this LTX 2.3 multi-image reference update works in practice. The video shows how multiple images can be imported into one LTX video workflow, how the 10-second similarity lock improves character stability, and how to run the workflow online without rebuilding the full ComfyUI environment locally.

This ComfyUI workflow is designed for LTX 2.3 multi-image reference video generation with stronger 10-second character consistency control. Its main purpose is to let creators import multiple reference images and use them as visual anchors across the video timeline, while reducing the common problems of identity drift, clothing changes, unstable facial features, and chaotic reference mixing.

The workflow is built around the LTX 2.3 distilled 1.1 route. It uses a Gemma3 fp8 text encoder, LTX Audio VAE, LTXVConditioning, LTX2_NAG, Seed Everywhere, ManualSigmas, CFGGuider, SamplerCustomAdvanced, LTXVLatentUpsampler, LTXVConcatAVLatent, LTXVSeparateAVLatent, VAEDecodeTiled, CreateVideo, and SaveVideo. Compared with a basic image-to-video graph, this workflow is structured as a production pipeline with multiple reference inputs, staged sampling, latent upscaling, and final high-definition refinement.

The key node is LTXVAddGuideMulti. This node allows multiple images to be inserted as timeline guides. In this version, the workflow uses a four-image reference guide structure, with each image assigned its own frame index and strength value. This means one image can define the starting character identity, another can guide the pose or clothing, another can carry a later-frame visual state, and another can stabilize environment or composition. The workflow also keeps an expandable guide structure, making it easier to extend reference control when needed.

The major update is the 10-second similarity lock strategy. The workflow includes 10S Sigmas Easing and similarity / anchor-style guidance settings across the staged sampling process. Instead of letting reference control collapse after the first few frames, the workflow keeps reference influence more stable through the generation window. This is important for character videos, because the face, hairstyle, outfit, body proportions, and scene logic often drift during longer clips.

The generation process is divided into three stages. The first stage builds the initial composition and reference relationship. The second stage performs latent-space upscaling while applying stronger mid-stage similarity guidance and weak anchor control. The third stage focuses on high-definition refinement with lighter similarity control, improving detail while trying not to destroy the established identity.

The workflow also uses LTX2_NAG and a universal negative prompt system to suppress flicker, frame jitter, identity drift, subtitles, watermarks, logos, broken anatomy, unstable mouth shapes, and unwanted text. This makes the final output cleaner and more suitable for publishing.

Compared with ordinary multi-image video workflows, this version is more reliable for character-stable generation. It is suitable for AI character videos, multi-reference story clips, MV fragments, cinematic shots, product-style character demonstrations, Bilibili content, YouTube showcases, RunningHub publishing, and Civitai workflow sharing.

Main features:

• LTX 2.3 multi-image reference video workflow

• Major update for multi-image import

• Four active reference image guide structure

• LTXVAddGuideMulti timeline guide control

• Frame index control for each reference image

• Strength control for each reference image

• 10-second similarity lock strategy

• 10S Sigmas Easing for staged control

• Similarity and weak-anchor stability logic

• LTX2_NAG universal negative guidance

• Three-stage rendering structure

• LTXVLatentUpsampler high-resolution transition

• AV latent concatenation and separation

• VAEDecodeTiled and final video output

Suggested workflow:

Prepare several clean reference images first. Use one image for the main character identity, one for clothing or pose, one for a later-frame visual direction, and one for scene or composition support. Load the images into the workflow and check each frame index and strength value before rendering. Start with a short test to confirm whether the references are being fused correctly. If the character changes too much, reduce prompt conflict and keep the 10S similarity / anchor settings active. If one reference dominates too strongly, lower its guide strength or move its frame index. After the first-stage composition is stable, continue into latent upscaling and final HD refinement.

⚙️ RunningHub Workflow

Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2061671863138476034?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/BV1nVVr6QEd8/

☕ 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/2061671863138476034?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/BV1nVVr6QEd8/

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。