Free workflows are available on the Pixaroma Discord server in the pixaroma-workflows channel discord.gg/gggpkVgBf3 You can now support the channel and unlock exclusive perks by becoming a member: pixaroma ruclips.net/channel/UCmMbwA-s3GZDKVzGZ-kPwaQjoin Check my other channels: www.youtube.com/@altflux www.youtube.com/@AI2Play
Great tutorial, works exactly as described. As you mentioned the first few clips are a bit off and improve with further iterations. Looking forward to seeing how the technology improves! Thanks for posting.
Wow bro! It generated on my RTX 3060 12gb in less than 2 minutes, less time than when I generate on a paid platform I used. Now, I just have to find out how to use a 4k upscale node and I'm done. Bro, you just made me save lots of money. Really good for a poor Brazilian guy.
This is impressive! 🎥✨ Local video generation models like LTX could truly shape the future of open-source AI, making creativity more accessible to everyone.
I love your videos. When you started on Episode 01 I put comfy on a 125 GB SSD, now I had to move everything to an 8 TB SSD I was running out of space.
i wonder if is possible to use a singel image and text describe to generate a AI video and every feames can be used in unity 2d animation frame sequence
Maybe you can get the frames from the video but not sure is so advanced to do animation like that, i mean i got with kling ai advanced animation but ltx wasn't quite so advanced
Good. Let's wait to see if the video can be looped properly. So far this is the first step, but the way it always cuts off abruptly, a lot of things are not available yet.
Sir, which python version is required for ComfyUI? Also, when I install LTX video from the manager, it says (IMPORT FAILED). What could be the reason for this?
I installed the portable version so it install the version it needs and its own environment , i have also updated to this maybe it helps, first i uninstalled then installed, I got Name: torch Version: 2.5.1+cu124 Name: torchvision Version: 0.20.1+cu124 Name: xformers Version: 0.0.28.post3 Name: torchaudio Version: 2.5.1+cu124
Hello and thank you.. One question. On the ComfyUI_examples page, its saying to download t5xxl_fp16.safetensors to /models/text_encoders folder. However you say to put that file in the /models/clip folder. Can you assist?
I think it works on both folders try one if not copy them in other, normaly is a text encoder so it should go to text encoder. I think i got used to put on clip because there i put for flux
Thank you for the tutorial! It's one of the best ones when it comes to local video diffusion. Is your audio AI generated as well? Some words sound suspicious. :))
I did update all and installed video helper but it still doesn't see these LTVX nodes. I do have the samplercustom node and video combine nodes. I also tried installing LTXVideo and LTXTricks but this still didn't give me those nodes. LTXVConditioning LTXVScheduler LTXVImgToVideo
For image to video I get this error: - Required input is missing: noise_scale Do I need a "get image size" node in between load image and LTX config nodes? I can't seem to find a get image size node in my node manager.
Try one of the workflows that i tested and worked from my discord ok pixaroma-workflows channel are free, so you can check to see if works and then you can modify that
VAE load device: cuda:0, offload device: cpu, dtype: torch.float32 no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded. ?
are you having enogh vram? that might be the cause, if you have model and clip it should work if you a powerful nvidia card with this 2 models Download ltx-video-2b-v0.9.safetensors into models/checkpoints folder huggingface.co/Lightricks/LTX-Video/tree/main Make sure t5xxl_fp16 is in your models/clip folder huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/tree/main/text_encoders
So i followed all your step by steps to generate a generic video of a woman walking (basically in the prompt i wrote: a woman walking) as simple as that, negatives were: blurry, noise, deformed. I went with a very quick prompts just to test it out. Video came out horribly terrible, it's pretty much as if 20 buckets of paints were tossed in the air and splashed on the ground. that's the video i got. i tempered a little with the cfg etc... (You know, self exploit) but it didn't get worse , the video was still (the paint example lol) with extremely minor improvement however nowhere a very clear / detailed / high res video of a woman walking . Any tips / ideas/ or suggestions? thank you
try to use long detailed prompts generated with an LLM it expects long prompts with details, simple prompts will fail more than longer detailed prompt. I use chatgpt for example to get detailed long prompts
I use voiceair, they have the voice from elevenlabs, I can do a video for the ai2play channel maybe in the future, just need to find some time, I have so many projects scheduled :)
Thank you very much for this amazing tutorial and the workflows shared. This was my first attempt at generating video, and everything worked flawlessly (and even faster than I expected !!!) I have a couple of questions, in case you’re up for answering them: (Silly question) Img2Vid - Can the workflow be set up to generate only the MP4 video without saving the PNG reference image to the output folder? (Complex question) Img2Vid - Do you know of any workflow that allows specifying both the start and end frames of the video? That would be super useful for me... Thanks so much for your hard work!
I dont know any workflow yet, yeah that with end frame will be useful. For saving I dont know why save the png also. I tried to disable in video combine the output, but then it doesnt save png and no video, so only if you right click and save it from there is like a preview
@@ВладиславАндреев-о7о you can try to install this comfyui version on another folder, it install automatically all the nodes i used so far github.com/Tavris1/ComfyUI-Easy-Install made by a member of our discord community
Missing Node Types: LTXVScheduler LTXVConditioning EmptyLTXVLatentVideo - ComfyUI updated, VideoHelper installed but there is no way to install these missing nodes in your workflow. Do not show up in the manager.
If you click on manager and install missing nodes still doesn't appear? Maybe update comfyui from the update folder there is a bat there, not sure why it does appears to you
it should not take that long, not sure why, try with less length like 97 and that 768x512px and see if all the times is that speed or only the first time. I asked someone and said on 12gb of vram takes like under 2 minutes, and depends on the text clip encoder also, the fp16 worked better for some, for other fp8 was faster etc, so try other clips maybe
😂 I am using Ai to convert my text to audio, so it reads different sometimes, depending on the generation. I think it didn't read right and i gave the numbers in words, maybe I didn't give the text right since English is not my native language
Why is this shit so goddamn archaic? Every single video I watch there is some features in the video that I don't have. At 1:16 you're in a manager window downloading a video helper and I don't have those screens. smdh. I'm about to give up.
can you post a screnshot in the comfyui channel on my pixaroma discord server? you should have all the things I have. I will try to help or someone from discord will help, Just trying to understand what it works and not, and what you get. Did you installed manager?
![Shoot EPIC MOVIES with this FREE AI Tool! [ComfyUI Tutorial + Free Workflow]](http://i.ytimg.com/vi/gHI6PjTkBF4/mqdefault.jpg)
Free workflows are available on the Pixaroma Discord server in the pixaroma-workflows channel discord.gg/gggpkVgBf3
You can now support the channel and unlock exclusive perks by becoming a member:
pixaroma ruclips.net/channel/UCmMbwA-s3GZDKVzGZ-kPwaQjoin
Check my other channels:
www.youtube.com/@altflux
www.youtube.com/@AI2Play
Great tutorial, works exactly as described. As you mentioned the first few clips are a bit off and improve with further iterations. Looking forward to seeing how the technology improves! Thanks for posting.
The quality of the generated videos is very decent. Thank you very much!
Yeah I can work with some of those, image to video and some good prompts and a lot of variations 😁
Wow bro! It generated on my RTX 3060 12gb in less than 2 minutes, less time than when I generate on a paid platform I used. Now, I just have to find out how to use a 4k upscale node and I'm done. Bro, you just made me save lots of money. Really good for a poor Brazilian guy.
Did you figure out how to upscale to 4k? I thought the video was fantastic, I have a really old GTX 1080! And you managed to create it in 10 minutes!
This is impressive! 🎥✨ Local video generation models like LTX could truly shape the future of open-source AI, making creativity more accessible to everyone.
thanks Uday :)
Thank for this pixaroma! Liked and subscribed! And I'll watch the whole playlist if I can!
Thank you ☺️
I love LTX video, especially with the enhanced output nodes.
I love your videos. When you started on Episode 01 I put comfy on a 125 GB SSD, now I had to move everything to an 8 TB SSD I was running out of space.
There so many cool things to try so more fun more space it needs 😂 I am almost out of space also, I have to see what things I can delete
cant wait to watch yet another ai video for ai videos
Thanks for sharing. Looking forward to implementing it.
always love your videos. Keep them coming :)
thank you 🙂
Great video and series, thanks for all your hard work. I started playing with the samplers, getting much better results with ddim.
Great to hear 🙂
Awesome! Super helpful. Thank you! Subbed
Interesting stuff!
Thanks!
nicely done overview of the model!
Great video, great model, great WF. Now we have to find a way to continue the video, because 10 seconds max is short.😉
You could take the last frame of the video and use it to make another video
that's what I do, but sometimes it goes in the wrong direction.@@pixaroma
Unless someone make a custom node or something or a better model i don't know a fix
Awesome I was at your discord server when I saw Video Generation finally I was waiting for this Thanks a lot 🎉❤
Thanks 🙂
another great tutorial,thank you keep going )
thanks 🙂
It ran on my M3 MacBook in less than 2 minutes😮
I assume is good
@ you’re awesome!
Thanks for sharing!
Really interesting tutorial. Thank you.
Thanks Marcel ☺️
thank you, great tutorial
i wonder if is possible to use a singel image and text describe to generate a AI video and every feames can be used in unity 2d animation frame sequence
Maybe you can get the frames from the video but not sure is so advanced to do animation like that, i mean i got with kling ai advanced animation but ltx wasn't quite so advanced
@@pixaroma i will try your advices, if i can't make it i hope it will becom your EP26
Great..thanks for sharing!!
Good. Let's wait to see if the video can be looped properly. So far this is the first step, but the way it always cuts off abruptly, a lot of things are not available yet.
Sir, which python version is required for ComfyUI? Also, when I install LTX video from the manager, it says (IMPORT FAILED). What could be the reason for this?
I installed the portable version so it install the version it needs and its own environment , i have also updated to this maybe it helps, first i uninstalled then installed, I got
Name: torch
Version: 2.5.1+cu124
Name: torchvision
Version: 0.20.1+cu124
Name: xformers
Version: 0.0.28.post3
Name: torchaudio
Version: 2.5.1+cu124
Hello and thank you.. One question. On the ComfyUI_examples page, its saying to download t5xxl_fp16.safetensors to /models/text_encoders folder. However you say to put that file in the /models/clip folder. Can you assist?
I think it works on both folders try one if not copy them in other, normaly is a text encoder so it should go to text encoder. I think i got used to put on clip because there i put for flux
Thank you!
Thank you for the tutorial! It's one of the best ones when it comes to local video diffusion.
Is your audio AI generated as well? Some words sound suspicious. :))
Yeah all the audio is generated from the text i give it, so is not always perfect but it does the job
what are th ui imporvments you use, like, ram, vram gpu usage and the progress bar?
I just installed the crystools node from manager, custom nodes manager, and those appear where you bar is, if is on top appears there
@pixaroma thank you
I did update all and installed video helper but it still doesn't see these LTVX nodes. I do have the samplercustom node and video combine nodes. I also tried installing LTXVideo and LTXTricks but this still didn't give me those nodes.
LTXVConditioning
LTXVScheduler
LTXVImgToVideo
Apparently update all didn't work completely, had to do a git pull on the folder.
Sometimes updating from manager fails
i got the same issue, can you help me out please.
For image to video I get this error: - Required input is missing: noise_scale
Do I need a "get image size" node in between load image and LTX config nodes? I can't seem to find a get image size node in my node manager.
Try one of the workflows that i tested and worked from my discord ok pixaroma-workflows channel are free, so you can check to see if works and then you can modify that
VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded. ?
are you having enogh vram? that might be the cause, if you have model and clip it should work if you a powerful nvidia card with this 2 models
Download ltx-video-2b-v0.9.safetensors into models/checkpoints folder
huggingface.co/Lightricks/LTX-Video/tree/main
Make sure t5xxl_fp16 is in your models/clip folder
huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/tree/main/text_encoders
So i followed all your step by steps to generate a generic video of a woman walking (basically in the prompt i wrote: a woman walking) as simple as that, negatives were: blurry, noise, deformed. I went with a very quick prompts just to test it out. Video came out horribly terrible, it's pretty much as if 20 buckets of paints were tossed in the air and splashed on the ground. that's the video i got. i tempered a little with the cfg etc... (You know, self exploit) but it didn't get worse , the video was still (the paint example lol) with extremely minor improvement however nowhere a very clear / detailed / high res video of a woman walking . Any tips / ideas/ or suggestions? thank you
try to use long detailed prompts generated with an LLM it expects long prompts with details, simple prompts will fail more than longer detailed prompt. I use chatgpt for example to get detailed long prompts
great stuff! What are you using for your AI voice? It sounds great. In the future can you make a tutorial on how to create an AI voice like that?
I use voiceair, they have the voice from elevenlabs, I can do a video for the ai2play channel maybe in the future, just need to find some time, I have so many projects scheduled :)
@@pixaroma awesome thanks!
@@pixaromaDoes it just translate what you say in another voice? Or are you writing it down
I write it in english and give that text to get audio
Great job, thanks ! How about a tutorial about Hunyuan video ?
The license dont allow use of it in Europe so i can not do a video about it
@@pixaroma A shame :( Thanks for your answer ;)
Thanks a lot!!!!!!!!!!!!!!
Thank you very much for this amazing tutorial and the workflows shared. This was my first attempt at generating video, and everything worked flawlessly (and even faster than I expected !!!)
I have a couple of questions, in case you’re up for answering them:
(Silly question) Img2Vid - Can the workflow be set up to generate only the MP4 video without saving the PNG reference image to the output folder?
(Complex question) Img2Vid - Do you know of any workflow that allows specifying both the start and end frames of the video? That would be super useful for me...
Thanks so much for your hard work!
I dont know any workflow yet, yeah that with end frame will be useful. For saving I dont know why save the png also. I tried to disable in video combine the output, but then it doesnt save png and no video, so only if you right click and save it from there is like a preview
Sorry, i have missed LTXV node... which one i should install?
I had update ComfyUI from manager
I have included in the video description all the details, with links
i have some issues with Viideo node, It is not installed
@@ВладиславАндреев-о7о you can try to install this comfyui version on another folder, it install automatically all the nodes i used so far github.com/Tavris1/ComfyUI-Easy-Install made by a member of our discord community
Can you add styles to the videos and other methods just img 2img
I didn't play too much with video, once i learn more things I can do more videos
Is there a limit on how long video you can generate?
I only did 5 seconds, i saw online some did with some extra nodes somehow 10 seconds but all ai seems to have a limit around 5-10 sec
Very nice. To bad you need a lot of resources. Do you know if we can create computers with more graphic cards ?
I don't know, but this model worked on 12gb video card
How to add node to save to video mp4 instead of webp? Also how to make video longer? Images distort with large prompt.
I showed in the video, i used that node from video helper, check the entire video
@@pixaroma Yes, I saw the new node. But the video result morphs and distorts most of the time. Any fix around for this?
@@bestof467 long prompts, slow motion, that improve it, all models have glitches, including Sora
Cannot find the workflow?
Is on discord on pixaroma-worflows channel
Arent there better video upscalers with better models than Topaz? For instance control the creativity or prompt option
it might be, but I dont know any, I used that one since I have it for a year or so
Can’t you use foolhardy upscaler 4K
Use the video combine node to save it as mp4. No one wants to save as webp format
I covered MP4 saving at minute 5:35 🙂
Missing Node Types:
LTXVScheduler
LTXVConditioning
EmptyLTXVLatentVideo
- ComfyUI updated, VideoHelper installed but there is no way to install these missing nodes in your workflow. Do not show up in the manager.
If you click on manager and install missing nodes still doesn't appear? Maybe update comfyui from the update folder there is a bat there, not sure why it does appears to you
@@pixaroma Nope - no way to install. Lots of users having the same issue, according to GitHub. This project seems to be broken 😕
👋 hi
Hi ☺️
4070ti takes about 5 minutes why?
it should not take that long, not sure why, try with less length like 97 and that 768x512px and see if all the times is that speed or only the first time. I asked someone and said on 12gb of vram takes like under 2 minutes, and depends on the text clip encoder also, the fp16 worked better for some, for other fp8 was faster etc, so try other clips maybe
you can also try to replace vae decode with vae decode tilled, might help with the memory
@7:01 "RTX four thousand ninety" huh? are you a robot? cuz the rest of us say forty-ninety.
😂 I am using Ai to convert my text to audio, so it reads different sometimes, depending on the generation. I think it didn't read right and i gave the numbers in words, maybe I didn't give the text right since English is not my native language
I am fast typist! I can type 5000 chars a min. The text is complete gibberish random garbage though... The same is with AI video nowadays...
I am sure it will get better in the future, like with AI images, we already can do cool images and it started really bad with a few pixels images.
Why is this shit so goddamn archaic? Every single video I watch there is some features in the video that I don't have. At 1:16 you're in a manager window downloading a video helper and I don't have those screens. smdh. I'm about to give up.
can you post a screnshot in the comfyui channel on my pixaroma discord server? you should have all the things I have. I will try to help or someone from discord will help, Just trying to understand what it works and not, and what you get. Did you installed manager?