If your system allows you to get the complete video in one loop, disable the rest and combine as well.

If you need more loop or less loop, don't forget to update the Combine Group image batch multi & audio concat accordingly.

I have setup the sampler & upscaler steps to 8 + 3 ( Not using manual sigmas to give you more freedom to choose )

Try to keep the loops in multiples of 24 to help you keep the audio duration simple.

Update the length, Split Images node to the number of frames you want out of one loop.

LTX 2.3 is weird with numbers, so I don't suggest you change any of the dimensions inside the Subgraphs if you don't want it to become a mess.

Feel free to experiment with the strength of loras & IC Lora guide strength. I have set it to the best combination I have experienced in past few days of testing.

Most of the nodes have set & get applied but some have been left intentionally to help you understand the workflow better if you want to change anything or add more loops.

LTX 2.3 first step is usually late, the actual time will only be realised by 3rd or 4th step which are faster.

You really need to find out the bottleneck of your system with LTX2.3 wrt to Resolution & Max Length you can generate. Always go one step lower where there's no bottleneck, you can queue workflows & leave them to do the job in the time of you estimated. This will take sometime to find out if you haven't already.

Motion Sync in LTX 2.3 for now seems to be heavier on the system than T2V/I2V where good length & good resolution can come out of one single batch of such loop.

If you want a double stage workflow ( introduces artifacts & instability with motion adherence ), enable the upscale group & output the SeparateAVlatent Video latent to VAE Tiled Decode & Audio latent to VAEAudio Decode. Enable Upscaler Lora. And deactivate in Multi lora.

You can also use LTXV Spatial Temporal Tiled Decode node to give you more refined output in either case ( slightly more time than VAE Tiled Decode ).

This workflow will work with both Dev & Distilled models, just replace the models with the ones you want to use. I have used models for LOW VRAM requirements.

1024 X 576 works best if your input image is good quality. You can use RTX Super Resolution or Upscale Model to enhance the video later or connect them with the output of your image batch multi node if you want it all at one go. Don't forget to attach the final concatenated audio in the create video node of the same.

Final note, Motion Sync in LTX 2.3 is not as accurate as in wan models. Same combinations can work sometime & sometime they won't give you good output. Still in work and depends on your IC motion control lora + IC lora guide strength combination a lot. You can let me know what worked better for you. So I can update it in the next version.

If you like this workflow, please support & give credits if you use it for make a better workflow.