A workflow uses the latest LTX2.3 Eros and LTX Anchor node to generate high-consistency long videos.
Other model details are within the workflow.
Try the workflow online for free: https://www.runninghub.ai/post/2057662125262331906
Please check the video tutorial before you use this workflow:
Description
FAQ
Comments (4)
hello, I love your workflow, thanks for sharing! Any plan for a image only to video+audio workflow? Or I'll try to add Kijay's anchor to my workflow but I'm a noob
It's in the workflow, just gota fix a few things: 1) Disable "Group" (it's the manual audio loader group). 2) Under First Sampling Group, lower left side there's a node called Empty Latent Audio. Connect that to the audio latent on the LTXVConcatAVLatent just to the right of it (replace the existing connection). 3) Under the User Inputs Group, there's a Math Expression node that will throw an error because the Get_audio_length is now disabled from step 1 above. Just add an integer box of some sort and connect to b for the number of seconds for the video/audio. I like the slider from the comfyui-mxtoolkit nodes.
Looks like a much cleaner implementation than the official WFs. I can confirm that using the TS anchor node does work reasonably well for preserving a likeness without help from additional LoRAs. Their official workflows are a real mess, but that node can definitely be dropped right into any existing WF, even without the additional guiders and amplifiers. Slow though. I'm not a fan of tiled sampling node, but I suppose if one must tile, might as well go whole-hog.
Great workflow, thanks for sharing!