LTX 2.3のClean Plate IC-LoRAをお試し✨ 動画内の人や車などをマスクなしで消し、隠れていた背景を補完し、何もいなかったように作り直す 渋谷スクランブル交差点から始めて、複数素材でどこまで消せるか検証しました🧐 動画素材サイトを利用し、投稿用にモザイク処理をしました #LTX23 #ComfyUI

LTX-2.3-22b-IC-LoRA-Clean-Plate はこちら👇

This is cool. During 2020 there was yt channel where man was walking around the city showing empty streets. Somehow he mamaged to remove all people in postprocessing. That reminds me of that channel.

Thank you! I didn’t know about that 2020 channel, sounds cool.

I found it! He claimed he woke up in a hospital in the year 2027 to find humanity had completely vanished

Thank you for sharing the link! That's a really interesting concept. The way he made the streets look completely empty is impressive. Our LTX demo is trying to achieve something similar but with AI in one step. Cool to see the old-school approach too!

We imagine this is how the next ‘I am legend’ will be filmed 😎🤣

Erasing Shibuya crossing without a mask, that background fill looks seamless

Erasing Shibuya crowds without masks is a huge time saver.

これは恐れ入りました、 あっ、おはようございます☀ こんな事が出来たら最高👍️

AIライフハックさん、おはようございます！☀ なかなか面白いLoRAでした☺️ ローカルでサクッとこうゆう事が出来る時代ですね。 人だらけの渋谷の実際の動画からすべての人を削除し、実背景を使用した動画生成とかやれますね。

マジで周回遅れ感がありすげて焦り気味です💦 やっぱり触ってみないと分かりませんね。 何とかせねば...

Amazing work! The train clean plate looks flawless—how many tries or parameter tweaks did it take to get the inpainting this clean?

I didn't try many times for the train scene either. If you check the official Hugging Face page and follow the LoRA instructions and prompt, it works pretty straightforward without much tweaking. This LoRA is really capable.

Thats pretty awesome, I need to test that out!

Thank you! Hope you like the results when you test it 🙏

これはとんでもなくクールだ😌

ここから特定の者や人物だけピックアップして表示できるようになるのも時間の問題ッスね

Looks awesome! I'm curious about resolution...

Thank you! I generated it at 1280x704 (LTX 2.3 doesn't support 1280x720 directly), then combined the original and result vertically into 1280x1440. The LoRA worked really well even at that resolution.

おおスゲエ(･∀･)!! こういう使い方もあるわけですねえ｡😃

これって広告とかポスターの中の人物も消えますか?

wow wow

That like an inverse of the corridor key (@CorridorDigital). Looks cool

@CorridorDigital Thank you! Yeah, it's kind of the inverse of Corridor Key effect. Glad you liked it 🙏

Cool! Thats neat

这太酷了

おそるべし🤔

このLoRAはかなり凄いです！🙌人間や犬、車など次から次へと動画から消しまくりです☺️おそるべしです！

ほんと、やばいわね🤪

What the source footage limitations?

The main limitation is that it works best when enough of the background is visible around the subject. Very small distant people may remain, while large foreground subjects, such as a close-up face covering much of the frame, may not be removed at all. Results also vary depending on scene complexity, motion, and how much background the model can reconstruct.

@ComfyUI is there a way we can get this in comfy cloud?

Javier Único Sobreviviente really come from AI27

💀

Noice, what's the input resolution limit?

Thank you! I used 1920x1080 source and resized to 1280x704 in ComfyUI. It worked very well.

What's the upscale model used to go back to 1080p-4K?