Skip to content

Conversation

@TTPlanetPig
Copy link

@TTPlanetPig TTPlanetPig commented Apr 20, 2025

by simply replace the last predicted image by inserted end image. now we can easily control the video to what we want. the Pincer attack from TeneT now is complete.
Thanks my friend "hhy" helped me on the coding as I only have the idea but he is much better on coding!

4.20.-1.mp4

image

250420_204753_497_5473_10.mp4

@FragmentY
Copy link

nice work!

@luze1011
Copy link

NICE

@FurkanGozukara
Copy link

Amazing work

I have implemented into my app and working amazing

My app tutorial : https://youtu.be/HwMngohRmHg?si=ImYvFey-R030fbiM

screencapture-localhost-7860-2025-04-20-20_19_26

250420_194342_833_7054_seed269740392_37.mp4

screencapture-localhost-7860-2025-04-20-20_25_31

@kabachuha
Copy link

kabachuha commented Apr 20, 2025

Interpolating between the start and the end conditionings depending on the frame number can be cool as well (actually, such interpolation is a unique feature of next frame prediction models), it will require some caching though, but it's for the future maybe :)

Anyway, amazing work, works like a charm! 🚀

@rkfg
Copy link

rkfg commented Apr 20, 2025

I tried doing the same and if the images are different enough the effect is bad. All segments but the first would animate the last frame, and the first segment would quickly turn from the first frame to last. Maybe I did something wrong, can you try with a longer video like 10-15 seconds?

@zadokov
Copy link

zadokov commented Apr 20, 2025

It could have been great but for now this only works for the first 1 second (all the transformation is done on 1st second and all the other seconds are just static).

@lllyasviel
Copy link
Owner

I think maybe I should create a repo like FramePackSoftware or FramePackAdvanced or some similar names as a independent repo to merge and implement ideas. This main repo is a research repo and need to be made simple. I will probably move this PR. Let me think how to process it

@zadokov
Copy link

zadokov commented Apr 20, 2025

I think maybe I should create a repo like FramePackSoftware or FramePackAdvanced or some similar names as a independent repo to merge and implement ideas. This main repo is a research repo and need to be made simple. I will probably move this PR. Let me think how to process it

@lllyasviel - All Issues and PRs here seem related to the software/studio - Maybe create a new repo for the Research?

@rkfg
Copy link

rkfg commented Apr 20, 2025

It could have been great but for now this only works for the first 1 second (all the transformation is done on 1st second and all the other seconds are just static).

That was my experience as well, I described my findings here: #32 (comment)

@spawner1145
Copy link

Amazing work

I have implemented into my app and working amazing

My app tutorial : https://youtu.be/HwMngohRmHg?si=ImYvFey-R030fbiM

screencapture-localhost-7860-2025-04-20-20_19_26

250420_194342_833_7054_seed269740392_37.mp4
screencapture-localhost-7860-2025-04-20-20_25_31

I have seen you in many issues and you are always trying to copy others' open source ideas to make money,don't you feel ashamed?

@TTPlanetPig
Copy link
Author

I tried doing the same and if the images are different enough the effect is bad. All segments but the first would animate the last frame, and the first segment would quickly turn from the first frame to last. Maybe I did something wrong, can you try with a longer video like 10-15 seconds?

for long video, you will need more middle frame or change the schedule method. still under study...

@GJohGJ
Copy link

GJohGJ commented Apr 21, 2025

Nice work!! If we could use controlnet also, this make a difference in guiding from a existing video

total_generated_latent_frames = 0

latent_paddings = reversed(range(total_latent_sections))
# 将迭代器转换为列表
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

English comment would be great

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

中文评论才是精华

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

佩服😁

@andorxornot
Copy link

it would be cool to set keyframes for any frame, not just start/end.

@Eikwang
Copy link

Eikwang commented Apr 22, 2025

效果非常好!

@deniss-muhla
Copy link

it would be cool to set keyframes for any frame, not just start/end.

If we have start and end frames when we can just split video generation in sections...
Actually it's even better to have a separate generation for each part as it will be easier to describe motion.

IMHO the next step will be batch generation and joining.
For now I join them in the video editor.

@poprice29
Copy link

Amazing work

I have implemented into my app and working amazing

My app tutorial : https://youtu.be/HwMngohRmHg?si=ImYvFey-R030fbiM

screencapture-localhost-7860-2025-04-20-20_19_26
250420_194342_833_7054_seed269740392_37.mp4

screencapture-localhost-7860-2025-04-20-20_25_31

You are literally saying you implemented someone else's work and are essentially selling it. Have you no shame?

@mocat15
Copy link

mocat15 commented Apr 25, 2025

Great job, @TTPlanetPig and hhy! It worked great for me. One issue that I had was that the size of the output videos was smaller than the i2v default output files with the same settings. I turned down the MPEG compression to zero, which made me feel a bit better.

Thank you, @lllyasviel, for Framepack!

@CKVNZ
Copy link

CKVNZ commented Apr 27, 2025

Sorry im a new guy here. How do i implement this first-last frame into my already existing Framepack?
The current Framepack consume too much space with duplicate output & temp folder in C drive.

@julia-0105
Copy link

My first and last frame sampling stops after 2-3 times, and no video is generated. Is there anyone who has the same problem as me? How can I solve it?

@soliderwang
Copy link

That's great! It would be perfect if it supported 20 series graphics cards

@IckeBins67
Copy link

How to implement this? I do git pull, but it says already up to date.

@uselessgithub
Copy link

uselessgithub commented Apr 29, 2025

How to implement this? I do git pull, but it says already up to date.

You have to pull this specific PR 167, something like this over your local pull:
git fetch origin pull/167/head:pr-167
git checkout pr-167

a-r-r-o-w added a commit to huggingface/diffusers that referenced this pull request Apr 29, 2025
uselessgithub added a commit to uselessgithub/FramePackPlus that referenced this pull request Apr 29, 2025
@jkpwed
Copy link

jkpwed commented May 3, 2025

I don't know python but I can see this opens the doors to Slow Motion, and Frame Interpolation (in between frames) converting 30 fps video to 60 (or higher).

@tin2tin
Copy link

tin2tin commented May 3, 2025

Has anyone succeeded in using video as more start-frames and end-frames? If possible, FramePack could be used for video inpainting with motion trajectories intact.

a-r-r-o-w added a commit to huggingface/diffusers that referenced this pull request May 6, 2025
* add transformer

* add pipeline

* fixes

* make fix-copies

* update

* add flux mu shift

* update example snippet

* debug

* cleanup

* batch_size=1 optimization

* add pipeline test

* fix for model cpu offloading'

* add last_image support; credits: lllyasviel/FramePack#167

* update example with flf2v

* update penguin url

* fix test

* address review comment: #11428 (comment)

* address review comment: #11428 (comment)

* Update src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video_framepack.py

---------

Co-authored-by: Linoy Tsaban <[email protected]>
@SimmC2
Copy link

SimmC2 commented May 7, 2025

did have F1 ver?

@wozniakty
Copy link

Tried this out and the video started immediately with the endframe?

@IckeBins67
Copy link

works fantastic! thanks! i would have loved to integrate the slider for the resolution here, but i have no idea how.

@spawner1145
Copy link

works fantastic! thanks! i would have loved to integrate the slider for the resolution here, but i have no idea how.

Maybe you can have a look at my code? I've integrated some of the most commonly used functions.
https://github.com/spawner1145/sd-webui-framepack.git

@IckeBins67
Copy link

works fantastic! thanks! i would have loved to integrate the slider for the resolution here, but i have no idea how.

Maybe you can have a look at my code? I've integrated some of the most commonly used functions. https://github.com/spawner1145/sd-webui-framepack.git

Thanks, but my Chinese is just as bad as my programming skills. Do you have a step by step guide?

@spawner1145
Copy link

works fantastic! thanks! i would have loved to integrate the slider for the resolution here, but i have no idea how.

Maybe you can have a look at my code? I've integrated some of the most commonly used functions. https://github.com/spawner1145/sd-webui-framepack.git

Thanks, but my Chinese is just as bad as my programming skills. Do you have a step by step guide?

I've already created a bilingual version of the README.

@QuadTrillionaire
Copy link

Hello everyone, I need help please.

I have used the install package of the original framepack, which works great, as well as the F1 add on.
My issue is that I tried to install the video input add on, but I get a frustrating error:

" File "C:\framepack_cu126_torch26\webui\demo_gradio_f1_video.py", line 16, in
import decord
ModuleNotFoundError: No module named 'decord' "

I then used pip install decord within the main directory where framepack was installed and ran into this error:

"PS C:\framepack_cu126_torch26> pip install decord
Requirement already satisfied: decord in c:\python312\lib\site-packages (0.6.0)
Requirement already satisfied: numpy>=1.14.0 in c:\python312\lib\site-packages (from decord) (2.2.2) "

Can someone please guide me out of this situation?

@rafa11rodrigues
Copy link

I think maybe I should create a repo like FramePackSoftware or FramePackAdvanced or some similar names as a independent repo to merge and implement ideas. This main repo is a research repo and need to be made simple. I will probably move this PR. Let me think how to process it

Hi
Did you create this repo? If not, do you plan to merge these extensions into the main branch?
I know we can pull from the specific PRs, but it would be very nice to have these features "natively" from the main official repo, or from a repo maintained and reviewed by the owner of the project

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.