Workflow: T2V R2V - Create consistent character videos from a reference face image with Best Face ID lora
T2V R2V - Create consistent character videos from a reference face image with Best Face ID lora
Use a close-up portrait image to create consistent videos from the reference image with the Face ID lora from @Alissonerdx
Easily create multiple scenes with same character. Since only the face image is the reference, you can prompt the scene, clothing, sequence of action as you desire ;- )
Get the lora here: https://huggingface.co/Alissonerdx/LTX-Best-Face-ID
Required node: https://github.com/alisson-anjos/ComfyUI-BFSNodes
Optional ID-Lora (audio) for the workflow with both: https://huggingface.co/Comfy-Org/ltx-2.3/tree/main/split_files/loras (TalkVid or CelebHQ - both are ok)
The workflow come in multiple variants
- Regular Face ID workflow just using image input
- Face ID with custom audio - with both face image and a custom audio mp3 as audio source.
- Face ID with ID-Lora (audio) - use both face reference image, and a reference audio (5-10 seconds) and prompt both the audio and video.
Try out the workflows here:
https://huggingface.co/RuneXX/LTX-2.3-Workflows/tree/main/Multi-ref-character-sheet
(also some workflows at the repo of the lora)
Nice job again @RuneXX
Is it possible to incorporate this type of merged model into your R2V moviemaker, music maker or R2V long workflows?
It sounds like it could keep the identity better over multiple shots / scenes?
But it mentions: Requires the memory-multishot path to get the identity benefit — a plain single-shot graph will not exercise the memory bank.
LTX-2.3 echoVid + ltxAud — surgical merge
JoyAI-Echo's video/conditioning branch married to LTX-2.3-distilled-1.1's audio branch
https://huggingface.co/joeygambino/ltx23-echoVid-ltxAud-surgical
Is it possible to incorporate this type of merged model into your R2V moviemaker, music maker or R2V long workflows?
Had same thought, specially the longer "music video" workflow that generates as a single scene, it might benefit hugely ;-)
Will try that asap
Might also benefit mutli-shot, at least the face stays consistent... so if your clothing description is excessive, it should stay fairly consistent
(and also, it might be beneficial sometimes to not.. say the person starts the story indoors, and then go outdoors wearing different clothes etc.. )
Main challenge would be the voice, but if using extend video loop, and the audio is in the reference part of past video, the voice should also stay consistent (or use ID-Lora for audio)
Will check out that echoVid model ;-)