The animator is based on the frame interpolation workflow from Lightricks Github page.
This ComfyUI workflow allows you to take several photos of a subject and interpolate everything in-between the photos, creating an animation.
To install:
Non-GGUF edition:
Read the instructions on Civitai's LTX 0.9.5 page. Then download the photo animator from this page and open LTXV095_photo_animator.json in ComfyUI.
GGUF edition:
Download the GGUF of your choice from City96's Huggingface. Q8 will (just) fit on a 10GB graphics card with all models loaded, more comfortably on 12GB. Smaller-size files will fit better at a cost of possible artifacts and mistakes appearing in the animation.
Then download the VAE from City96's Huggingface.
Also download the T5 text encoder of your choice from City96's Huggingface. Smaller models should work as CFG is set to 1 on this workflow.
When you've done that, open LTXV095_photo_animator_gguf-edition.json in ComfyUI.
Instructions:
Using the workflow, add the photos in sequence at the bottom. You should then consider what frame rate the animation should run at. The LTX default is 25fps but is 60fps on this workflow for smoother animations, but will take longer. Change it to your needs. Remember to change the video output frame rate at the same time.
Then, think about the number of frames you want between each photo, in multiples of 8 (based on how LTX works). For 25fps and ~1 second between the first and second photo, use frame idx 0 for photo 1 and frame idx 24 for photo 2. For 60fps, frame idx 0 and frame idx 56 or 64.
The number of frames in the bottom left should be the last frame idx + 1. E.g. if it's 200, set the frames to 201.
Notes:
Two of the six photo nodes groups are bypassed by default. You can hold Ctrl + drag a selection with the mouse to enable them, and the same to bypass other photo node groups.
Try to focus on one subject moving. If there are more than one, you may want to split the animation into several parts and join them together in a video editor, as LTX can become confused. The pens animation uses two photos at a time for that reason.
Take plenty of photos, and think about what you want to achieve beforehand. It's better to have too many than too few as the lighting will have changed if you need more!
Long prompts are normally recommended with LTX, but might not be necessary. The pens animation uses no prompt.
MMAudio was used on each video separately, as my ComfyUI is unable load it. Search for workflows on Civitai for a suitable one.
IPNDM is the only sampler that works in testing without causing large amounts of artifacts.
I can only run the GGUF version on my system due to hardware limitations. The regular version should work though. LTX 0.9.6 has recently been released and might work even better, though no GGUF version is available at present.
Good luck and happy animating!