Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

More LTX IC LoRA experimentation - I’m currently training a LoRA for seamless transitions. I think this one’s close to being ready to ship 🚚💨

Ingi Erlingsson 🪄

17,114 subscribers

18,403 Aufrufe • vor 5 Monaten •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

35 Kommentare

Profilbild von Rafa Schwinger 🇻🇦

Rafa Schwinger 🇻🇦vor 5 Monaten

@ltx_model I wonder if you could use a n x n grid as the initial frame and make the lora understand all of the in between frames at once

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

@ltx_model Absolutely possible I think The training data would take some time to put together and it would need some serious GPU power to train, but very doable I think

Profilbild von Jacques Broquard

Jacques Broquardvor 5 Monaten

@ltx_model You're pushing the fuckin' limits mate.

Profilbild von rob - comfyui

rob - comfyuivor 5 Monaten

@ltx_model holy lora

Profilbild von D(AI)vid

D(AI)vidvor 5 Monaten

@ltx_model Way better than the first version man. Good job!

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

@ltx_model

Profilbild von makeitrad

makeitradvor 5 Monaten

@toyxyz3 @ltx_model Nice ⚡️🤘

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

Download the LoRA and @ComfyUI workflow here:

Profilbild von Qwaig

Qwaigvor 5 Monaten

@ltx_model Heady

Profilbild von Pandoras Parabox

Pandoras Paraboxvor 5 Monaten

@ltx_model Looking forward to the release

Profilbild von FAT JC

FAT JCvor 5 Monaten

@ltx_model

Profilbild von MΛX V꩜X (Max Vox)

MΛX V꩜X (Max Vox)vor 5 Monaten

@ltx_model

Profilbild von 𝚁𝚑𝚞𝚊𝚗 𝙿𝚊𝚐𝚊𝚗𝚘

𝚁𝚑𝚞𝚊𝚗 𝙿𝚊𝚐𝚊𝚗𝚘vor 5 Monaten

@ltx_model Please let us know when it’s out!!

Profilbild von ρŁ𝐀𝔰Ｍʘ

ρŁ𝐀𝔰Ｍʘvor 5 Monaten

@ltx_model My gpu is ready

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

@ltx_model We’re dropping it later today on the @thesystms account if all goes to plan

Profilbild von enigmatic_e

enigmatic_evor 5 Monaten

@ltx_model Smooth

Profilbild von biglootpig

biglootpigvor 5 Monaten

@ltx_model Super cool

Profilbild von Jose Marcial Portilla

Jose Marcial Portillavor 5 Monaten

@ltx_model So good!

Profilbild von Safety Marc

Safety Marcvor 5 Monaten

@ltx_model This looks really good!

Profilbild von neb

nebvor 5 Monaten

@ltx_model So cool

Profilbild von Purz.ai

Purz.aivor 5 Monaten

@ltx_model WILD

Profilbild von Xavier Climent

Xavier Climentvor 5 Monaten

@ltx_model Thks that's i'm waiting for!!

Profilbild von fabdream.ai

fabdream.aivor 5 Monaten

@ltx_model so good !

Profilbild von Armand

Armandvor 5 Monaten

@ltx_model ⏰

Profilbild von 𝗜𝗚𝗜 𝗠𝗲𝗱𝗶𝗮

𝗜𝗚𝗜 𝗠𝗲𝗱𝗶𝗮vor 5 Monaten

@ltx_model @ingi_erlingsson @ltx_model try doing it without so much Copyrighted materials

Profilbild von Slop-Swap

Slop-Swapvor 5 Monaten

@ltx_model Super impressive

Profilbild von Exigent Movie

Exigent Movievor 5 Monaten

@ltx_model

Profilbild von EsotericHustler

EsotericHustlervor 5 Monaten

@ltx_model this is so peak.

Profilbild von Heba AI

Heba AIvor 5 Monaten

@ltx_model link to the ic lora?

Profilbild von Christopher Fryant

Christopher Fryantvor 5 Monaten

@ltx_model Would be great if we could steer these with text prompting. Either way, they're still super smooth.

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

@ltx_model Yeah you can prompt them I generally use an LLM to do it as it’s quicker for testing, but you could absolutely do it manually Dropping the LoRA and workflow a little later on

Profilbild von Christopher Fryant

Christopher Fryantvor 5 Monaten

@ltx_model Awesome, looking forward to it. @ComfyUI would be great to get this on Comfy Cloud if possible.

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

@ltx_model @ComfyUI Already on it - it’ll be on my Comfy Cloud profile and I’ll share the link here too

Profilbild von Christopher Fryant

Christopher Fryantvor 5 Monaten

@ltx_model @ComfyUI Oh hell yeah. Is it possible to have a fully custom duration on these videos for that? I could see this being very handy for marrying two clips together if you could get really specific with the timing, down to the frame.

Profilbild von Ingi Erlingsson 🪄

Ingi Erlingsson 🪄vor 5 Monaten

@ltx_model @ComfyUI Yeah you set the duration of each clip + the total frames and the transition is the difference between the two Everything goes through the inference though, so it’s best to trim the last/first few frames from the clips and use that There will be tips in the WF too

Ähnliche Videos

I Forgot to post about this one here "Ungrade" another IC LoRA I released a few days ago, this one shows just how LTX with IC LoRA is re-defining what can be done or not in post-production.

I Forgot to post about this one here "Ungrade" another IC LoRA I released a few days ago, this one shows just how LTX with IC LoRA is re-defining what can be done or not in post-production.

Mohamed Oumoumad

24,525 Aufrufe • vor 6 Monaten

Wow I recommend EVERYONE interested in FLUX + LoRAs + Forge watch this. Super comprehensive tutorial on: 1. Getting started with Forge 2. Training LoRA with Fluxgym 3. Using the LoRA 4. Generating a virtual influencer with LoRA 5. Mixing LoRAs And more! I myself learned a lot.

Wow I recommend EVERYONE interested in FLUX + LoRAs + Forge watch this. Super comprehensive tutorial on: 1. Getting started with Forge 2. Training LoRA with Fluxgym 3. Using the LoRA 4. Generating a virtual influencer with LoRA 5. Mixing LoRAs And more! I myself learned a lot.

cocktail peanut

78,226 Aufrufe • vor 2 Jahren

A look at the video outpainting workflow in ComfyUI using LTX 2.3. 🎥 Instead of cropping your 16:9 footage for vertical, use this IC-LoRA setup to generate the missing space and keep your original frame. Shout out to Mohamed Oumoumad for their open source contribution. Link to workflow below! 👇

A look at the video outpainting workflow in ComfyUI using LTX 2.3. 🎥 Instead of cropping your 16:9 footage for vertical, use this IC-LoRA setup to generate the missing space and keep your original frame. Shout out to Mohamed Oumoumad for their open source contribution. Link to workflow below! 👇

ComfyUI

19,041 Aufrufe • vor 6 Monaten

SDR to HDR from ComfyUI, I've trained a LoRA over Qwen Edit 2011 based on the principle used in A research I've had the previlige to work on with Naomi Ken Korem for video HDR on LTX. Links for the LoRA and workflow (+ my fun grading node) below.

SDR to HDR from ComfyUI, I've trained a LoRA over Qwen Edit 2011 based on the principle used in A research I've had the previlige to work on with Naomi Ken Korem for video HDR on LTX. Links for the LoRA and workflow (+ my fun grading node) below.

Mohamed Oumoumad

59,885 Aufrufe • vor 5 Monaten

We just open-sourced 3DREAL ! A new render-to-real IC-LoRA for LTX-2.3 from fal: turn any 3D / CG / game render into high 3d render, photorealistic, film-quality video, keeping your exact composition and camera move using the first frame.

We just open-sourced 3DREAL ! A new render-to-real IC-LoRA for LTX-2.3 from fal: turn any 3D / CG / game render into high 3d render, photorealistic, film-quality video, keeping your exact composition and camera move using the first frame.

fal

217,065 Aufrufe • vor 3 Monaten

Trained a voice lora on Walken using just 6 clips as a test to see how well VoxCPM handles learning a voice. Then I just did audio + image = video with LTX 2.3. This TTS model feels extremely powerful, fully open sourced (Apache 2.0) More below 👇

Trained a voice lora on Walken using just 6 clips as a test to see how well VoxCPM handles learning a voice. Then I just did audio + image = video with LTX 2.3. This TTS model feels extremely powerful, fully open sourced (Apache 2.0) More below 👇

A.I.Warper

24,253 Aufrufe • vor 5 Monaten

A little taste of what’s to come… ✨ Think this one’s a keeper? Drop a 🔥 if you’re ready for more!

A little taste of what’s to come… ✨ Think this one’s a keeper? Drop a 🔥 if you’re ready for more!

Cannons

11,512 Aufrufe • vor 1 Jahr

Testing out this LTX 2.3 LoRA for removing objects blocking a shot. There’s still a bit of jitter when the face goes out of view, but coming from shooting weddings, I definitely would’ve tried using this to save shots when something was blocking my view.

Testing out this LTX 2.3 LoRA for removing objects blocking a shot. There’s still a bit of jitter when the face goes out of view, but coming from shooting weddings, I definitely would’ve tried using this to save shots when something was blocking my view.

enigmatic_e

23,605 Aufrufe • vor 4 Monaten

Not only do we have seamless transitions from you to your ship, we've also added a sensor mode for all players to indentify threats and nice loot! #gamedev #indiedev

Not only do we have seamless transitions from you to your ship, we've also added a sensor mode for all players to indentify threats and nice loot! #gamedev #indiedev

Filip Coulianos

223,477 Aufrufe • vor 3 Jahren

“For myself, I expect to be close to perfect. I think I’m at that level at this point.” My full interview with Joe Burrow after a spirited #Bengals practice on Inside Training Camp Live NFL Network:

“For myself, I expect to be close to perfect. I think I’m at that level at this point.” My full interview with Joe Burrow after a spirited #Bengals practice on Inside Training Camp Live NFL Network:

Tom Pelissero

141,445 Aufrufe • vor 1 Jahr

The Indian Air Force may get the first delivery of LORA missiles by year‑end. India signed a 2023 deal with Israel for this high‑precision, long‑range missile system range up to 400 Kilometres. Using LORA against Pakistan would be exciting.

The Indian Air Force may get the first delivery of LORA missiles by year‑end. India signed a 2023 deal with Israel for this high‑precision, long‑range missile system range up to 400 Kilometres. Using LORA against Pakistan would be exciting.

Baba Banaras™

108,736 Aufrufe • vor 1 Jahr

Joe Flacco is a Cincinnati Bengal. You don’t think I’m ready to root for him?! I was born for this.

Joe Flacco is a Cincinnati Bengal. You don’t think I’m ready to root for him?! I was born for this.

Michael Turner

108,880 Aufrufe • vor 1 Jahr

I think i’m ready to fall in love after directing this video. I’m a lover boy. 🥹

I think i’m ready to fall in love after directing this video. I’m a lover boy. 🥹

Perliks

36,695 Aufrufe • vor 5 Monaten

Found a hardcore LoRA workflow! Generate a PLY point cloud --> Adjust angle in editor -->Refine with Qwen-Image-Edit-2511 Gaussian Splash LoRA. It accurately restores complex perspective shifts. As shown in the demo, it handles 3D rotation and can even restore high-def details from blurry close-ups. Within a 45° range, the consistency is unmatched. Huge thanks to 大雄 for this contribution! 🫡

Found a hardcore LoRA workflow! Generate a PLY point cloud --> Adjust angle in editor -->Refine with Qwen-Image-Edit-2511 Gaussian Splash LoRA. It accurately restores complex perspective shifts. As shown in the demo, it handles 3D rotation and can even restore high-def details from blurry close-ups. Within a 45° range, the consistency is unmatched. Huge thanks to 大雄 for this contribution! 🫡

Tongyi Lab

1,493,613 Aufrufe • vor 9 Monaten

$"Bo Nix had a follow-up cleanup procedure after a fractured ankle and he’s still expected to be ready for the season.. I would expect Bo Nix to be ready for training camp or very close to training camp" ~ Ian Rapoport #PMSLive$

"Bo Nix had a follow-up cleanup procedure after a fractured ankle and he’s still expected to be ready for the season.. I would expect Bo Nix to be ready for training camp or very close to training camp" ~ Ian Rapoport #PMSLive

Pat McAfee

101,553 Aufrufe • vor 5 Monaten

As I promised yesterday, I'll briefly explain LoRA training and share a workflow I made so you can do it quickly. First, let me answer a very common question: 'Why train LoRAs when we have such advanced models?' Even though we have incredibly advanced models now (like NBP), we still can't always get them to do specific things we want. Simplest example: the spritesheet LoRA I made the other day. I generated 1000 images with Nano Banana and only 100 were what I wanted. The LoRA I trained using those 100 images gives me nearly 100% consistent results. Second point is cost and speed. With LoRA, we can cut costs by 4-5x. And while doing that, we're generating 4-5x faster. How many images do you need for a good LoRA? This depends on your LoRA's complexity. For example, when I training the spritesheet LoRA, even though I used 100 images, I didn't include buildings in the training data, so this LoRA doesn't work for buildings. So think about your LoRA's use cases and add examples for as many use cases as possible to improve quality. What are paired images and how to train LoRAs for image-editing? When training LoRAs for image editing on fal, we call each edit example paired images - one with _start suffix, one with _end suffix. For example, if you're training a background remove LoRA, the unedited original photo will be your '_start' image. The image with background removed will be the '_end' image. Simply put: images we want to edit or use as reference get _start, target images we want to achieve get '_end'. Important: save both images with the same name. Like image332_start.jpg and image332_end.jpg. This way the system knows which images pair together. What about training LoRAs for models with multiple image inputs? Same logic. We still use _start and _end suffixes, but with one difference. Since there are multiple input images, we can number them: _start, _start1, _start2. Example: start images, 1st image = Woman portrait (image35_start.jpg) 2nd image = Glasses photo (image35_start1.jpg) 3rd image = Hat photo (image35_start2.jpg) Output image = portrait of woman wearing glasses and hat (image35_end.jpg) Can we do more detailed captioning? Yes. Similarly, you can improve training quality by creating a txt file for each set with the caption inside. Example: create image35.txt and write: 'Recreate the image by putting the glasses from the second image and the hat from the third image on the woman in the first image.' What are Steps? How many should I use? What's Learning Rate? Steps determines how many times the model sees and processes your training data (your images). Each step, the model learns a bit more. But as steps increase, so does the risk of overfitting. So there's no real default. But for a simpler LoRA with 20 paired images, 1000 steps is ideal. Here's a metaphor for the Steps and Learning Rate relationship: Imagine you have a balloon. Our goal is to inflate it to the optimal size. Steps = How many times we blow into the balloon Learning rate = How hard we blow each time If we blow too softly, we need to blow many more times. If we blow too hard, we risk popping it quickly and can't reach optimal size. Of course training won't explode, but it won't work as intended because it wasn't trained optimally. Training's done, now what? Once training's complete, you'll have a safetensors file. Every model you train on fal has a LoRA inference endpoint. In that inference, add your safetensors file link to the LoRA url input, and you can use your LoRA. Thanks for the read! The workflow in the video: If I forgot anything, let me know in the replies.

As I promised yesterday, I'll briefly explain LoRA training and share a workflow I made so you can do it quickly. First, let me answer a very common question: 'Why train LoRAs when we have such advanced models?' Even though we have incredibly advanced models now (like NBP), we still can't always get them to do specific things we want. Simplest example: the spritesheet LoRA I made the other day. I generated 1000 images with Nano Banana and only 100 were what I wanted. The LoRA I trained using those 100 images gives me nearly 100% consistent results. Second point is cost and speed. With LoRA, we can cut costs by 4-5x. And while doing that, we're generating 4-5x faster. How many images do you need for a good LoRA? This depends on your LoRA's complexity. For example, when I training the spritesheet LoRA, even though I used 100 images, I didn't include buildings in the training data, so this LoRA doesn't work for buildings. So think about your LoRA's use cases and add examples for as many use cases as possible to improve quality. What are paired images and how to train LoRAs for image-editing? When training LoRAs for image editing on fal, we call each edit example paired images - one with _start suffix, one with _end suffix. For example, if you're training a background remove LoRA, the unedited original photo will be your '_start' image. The image with background removed will be the '_end' image. Simply put: images we want to edit or use as reference get _start, target images we want to achieve get '_end'. Important: save both images with the same name. Like image332_start.jpg and image332_end.jpg. This way the system knows which images pair together. What about training LoRAs for models with multiple image inputs? Same logic. We still use _start and _end suffixes, but with one difference. Since there are multiple input images, we can number them: _start, _start1, _start2. Example: start images, 1st image = Woman portrait (image35_start.jpg) 2nd image = Glasses photo (image35_start1.jpg) 3rd image = Hat photo (image35_start2.jpg) Output image = portrait of woman wearing glasses and hat (image35_end.jpg) Can we do more detailed captioning? Yes. Similarly, you can improve training quality by creating a txt file for each set with the caption inside. Example: create image35.txt and write: 'Recreate the image by putting the glasses from the second image and the hat from the third image on the woman in the first image.' What are Steps? How many should I use? What's Learning Rate? Steps determines how many times the model sees and processes your training data (your images). Each step, the model learns a bit more. But as steps increase, so does the risk of overfitting. So there's no real default. But for a simpler LoRA with 20 paired images, 1000 steps is ideal. Here's a metaphor for the Steps and Learning Rate relationship: Imagine you have a balloon. Our goal is to inflate it to the optimal size. Steps = How many times we blow into the balloon Learning rate = How hard we blow each time If we blow too softly, we need to blow many more times. If we blow too hard, we risk popping it quickly and can't reach optimal size. Of course training won't explode, but it won't work as intended because it wasn't trained optimally. Training's done, now what? Once training's complete, you'll have a safetensors file. Every model you train on fal has a LoRA inference endpoint. In that inference, add your safetensors file link to the LoRA url input, and you can use your LoRA. Thanks for the read! The workflow in the video: If I forgot anything, let me know in the replies.

ilker

15,192 Aufrufe • vor 8 Monaten

OK so my Minimax H3 character swap LoRA has been trained to 1000 steps, and it's promising! It seems to hold the original video much more consistently than without the lora, including the original audio. However it falls short in a few ways: - Drifts from source in longer videos - Camera cuts turn into weird zooms or repositioning - Facial expressions don’t match well in close-ups I trained it on an RTX 4500 on Runpod overnight for around ~$11 but will need more VRAM for experimenting to get that 15s consistency. Also need to add more examples of videos with edits to have better edit-matching... and more expression-matching examples.... I've posted lora and dataset on huggingface for y'all either way: If anyone wants to throw me some X bucks or lend me their GPU I'll train it on a RTX 6000 Pro next 😁

OK so my Minimax H3 character swap LoRA has been trained to 1000 steps, and it's promising! It seems to hold the original video much more consistently than without the lora, including the original audio. However it falls short in a few ways: - Drifts from source in longer videos - Camera cuts turn into weird zooms or repositioning - Facial expressions don’t match well in close-ups I trained it on an RTX 4500 on Runpod overnight for around ~$11 but will need more VRAM for experimenting to get that 15s consistency. Also need to add more examples of videos with edits to have better edit-matching... and more expression-matching examples.... I've posted lora and dataset on huggingface for y'all either way: If anyone wants to throw me some X bucks or lend me their GPU I'll train it on a RTX 6000 Pro next 😁

akatz

35,350 Aufrufe • vor 19 Tagen

Hey Y’all💚 Has been a longish day and I’m ready to unwind🙌🏻 Momma skipped the THC so more for me🤗 Let’s share this blunt and a few bongs rips together, if you wanna🥹 Cheers🔥💨💨💨💨💨💨 PPP💨💨🤏🏻😉

Hey Y’all💚 Has been a longish day and I’m ready to unwind🙌🏻 Momma skipped the THC so more for me🤗 Let’s share this blunt and a few bongs rips together, if you wanna🥹 Cheers🔥💨💨💨💨💨💨 PPP💨💨🤏🏻😉

*~♡~Erin~♡~*

15,643 Aufrufe • vor 4 Monaten

“This is why I love the UK. I’m currently being paid to watch a live football match.” — Nigeria Lady says.

“This is why I love the UK. I’m currently being paid to watch a live football match.” — Nigeria Lady says.

𝐀𝐬𝐚𝐤𝐲𝐆𝐑𝐍

404,069 Aufrufe • vor 7 Monaten