Would it be possible to add end frame like in original FramePack/167?

[FernandoConti]

Hi there @colinurbs what an amazing platform you have created!

The original FramePack was extended with an end frame feature (because it works backwards), as described here:

lllyasviel#167

Would it be possible to add this to your code or the "Generate (Original)" tab in FramePack Studio?

If yes, I was hoping it would have two advantages:

1. Using a start and end frame PLUS a LoRA would give more precise control
2. Using the same image as start and end frame would create perfectly loopable videos

Of course these are just hopes and dreams / assumptions. Do you think it would work like that?

Many thanks to you and your colleagues for all your hard work!

(By the way, I'm using the Pinokio-installed version.)

[FernandoConti]

Now that I've been reading up on FramePack video looping, I've come across this fork:

https://github.com/red-polo/FramePackLoop

that has implemented code to successfully create looping FramePack videos with LoRA support, without glitches/morphs/warps at the looping moment.

Maybe this could be added to FramePack Studio?

I love it that so many developers expand FramePack, but it seems FramePack Studio is the most complete. Wouldn't it be great if more developers would join in on that effort? :-)

[colinurbs]

@FernandoConti the new version released last night includes end frame support for the original model. If you set the start and ending frame to the same image it will generate a video that will loop well.

Let me know if this is what you were looking for!

[FernandoConti]

Wow! That was fast! :-) Thank you so much!!

[pidgeon777]

@FernandoConti the new version released last night includes end frame support for the original model. If you set the start and ending frame to the same image it will generate a video that will loop well.

Let me know if this is what you were looking for!

Will it be possible to add multiple frames in the middle of the video, and not just the last one? E.g. to switch from one scene to the other after x seconds, for example.

I was curious to know if this was a technically doable thing.

Or, maybe, define multiple frames, each with a specific description?

[pidgeon777]

EDIT: Just found out this:

#73

An idea could be to also optionally define an image for each segment.

[FernandoConti]

Hi @colinurbs ,

I've tested the looping with the same image as start and end frame and... nothing happens! :-)

Meaning that, even though I use a LoRA with specific motions, those motions don't show in the video. There's minimal to no movement, resulting in a loopable video in which (almost) no motion is visible.

It would be great if the LoRA would do its work as in a generation without an end frame. Would that be possible? Also, why doesn't it work?

And indeed @pidgeon777 , adding not only a prompt per segment but also an image per segment would be amazing, but no idea if FramePack is compatible with that (or it would simply be multiple generations into one, but that's not what you're asking I guess).

[colinurbs]

Try lowering the end frame influence. Image per segment is coming but probably won't be out for at least a month or so.

…

On Thu, May 22, 2025 at 11:47 AM FernandoConti ***@***.***> wrote: *FernandoConti* left a comment (FP-Studio/FramePack-Studio#123) <#123 (comment)> Hi @colinurbs <https://github.com/colinurbs> @pidgeon777 <https://github.com/pidgeon777> , I've tested the looping with the same image as start and end frame and... nothing happens! :-) Meaning that, even though I use a LoRA with specific motions, those motions don't show in the video. There's minimal to no movement, resulting in a loopable video in which (almost) no motion is visible. It would be great if the LoRA would do its work as in a generation without an end frame. Would that be possible? Also, why doesn't it work? And indeed, adding not only a prompt per segment but also an image would be amazing. — Reply to this email directly, view it on GitHub <#123 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAY6A6F3UYT3OWWLPZ4Q2KT27XWSTAVCNFSM6AAAAAB5P52URWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSMBRG4ZDQNJRGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[FernandoConti]

@colinurbs thank you so much for all these new developments!

Do you think FramePack eventually could have its own ControlNet-type infrastructure? Or is it technically not possible to expand the Hunyuan video model like that?

I'm just imagining the power of start/end frame + LoRAs + ControlNet :-))

[arledesma]

Do you think FramePack eventually could have its own ControlNet-type infrastructure? Or is it technically not possible to expand the Hunyuan video model like that?

I'm just imagining the power of start/end frame + LoRAs + ControlNet :-))

@FernandoConti, You might be better off asking @lllyasviel that question as they are both the authors of FramePack as well as ControlNet

Their original FramePack paper: https://arxiv.org/abs/2504.12626
Their original ControlNet paper: https://arxiv.org/abs/2302.05543

[julia-0105]

Try lowering the end frame influence.

Image per segment is coming but probably won't be out for at least a month
or so.
尝试降低结束帧的影响。

每个片段的图像正在制作中，但至少还需要一个月才能发布。
左右。
…

Thank you for your excellent contribution! I noticed that the current work utilizes a LoRA trained on HY. Would you happen to have experimented with training based on Wan 2.1 as well?