Thank you, I love it, demonstration of Hunyuan "video cloning" Lora on 4090.

[diStyApps]

Demonstration of Hunyuan "video cloning" Lora and do some edits to the cloned video on 4090
https://github.com/user-attachments/assets/e70d9622-baae-4159-8846-8e7e5dbe5a68

[comfyonline]

Hi, do you just use 1 video and then cut them into multiple clips for training?

[fuzzyfazzy]

That's probably the best example I have seen yet.

[cchance27]

WOW thats impresssive, did you share the dataset or lora anywhere yet? interested to see what the setup looked like for something like this

[thebollo]

@diStyApps Would you be willing to share your workflow? I'm especially interested to see how you so successfully mixed multiple LoRAs. Every time I try to do that with Hunyuan the result is bad.

[fuzzyfazzy]

Hi, do you just use 1 video and then cut them into multiple clips for training?

As a pure guess - split the video into smaller clips with max 33 frames.

Caption all videos taking into account key stuff like dress and dounut etc

Train all the videos

Using vid 2 vid - use the original video as driver so it captures the original motion and make your prompt changes.

[fuzzyfazzy]

@diStyApps Would you be willing to share your workflow? I'm especially interested to see how you so successfully mixed multiple LoRAs. Every time I try to do that with Hunyuan the result is bad.

You need to add the blocks bit to the lora and only use double blocks and deselect all single blocks.

[thebollo]

@fuzzyfazzy Thank you! That really seems to work! I don't know why, but nevertheless it does.

[fuzzyfazzy]

@fuzzyfazzy Thank you! That really seems to work! I don't know why, but nevertheless it does.

Your Joan Holloway post popped into my feed this morning - really good!

[diStyApps]

Hi, do you just use 1 video and then cut them into multiple clips for training?

As a pure guess - split the video into smaller clips with max 33 frames.

Caption all videos taking into account key stuff like dress and dounut etc

Train all the videos

Using vid 2 vid - use the original video as driver so it captures the original motion and make your prompt changes.

Sorry for late replay i forgot i have post it, i am working on tutorial for this.

There is no video driver it all text to video after training,

Hunyuan_clone_3.mp4

[RichGua]

Hi, do you just use 1 video and then cut them into multiple clips for training?

As a pure guess - split the video into smaller clips with max 33 frames.
Caption all videos taking into account key stuff like dress and dounut etc
Train all the videos
Using vid 2 vid - use the original video as driver so it captures the original motion and make your prompt changes.

Sorry for late replay i forgot i have post it, i am working on tutorial for this.

There is no video driver it all text to video after training,

Hunyuan_clone_3.mp4

The stability looks great, look forward to your tutorial.