LTX-2 19B — COMPLETE COMMUNITY TOOLKIT
Everything you need to start creating with LTX-2 19B in ComfyUI, gathered in one community-maintained Civitai hub: base models, GGUF and FP8 variants, text encoders, VAEs, LoRAs, upscalers, ControlNet resources and ready-to-use workflows.
IMPORTANT
LTX-2 was created by Lightricks. I am not the original creator of the base model. This page mirrors and organizes useful public files from the official Lightricks ecosystem while also hosting my own workflows and community resources.
Official source: https://huggingface.co/Lightricks
START HERE — CHOOSE ONE MAIN MODEL
You do not need to download every file on this page.
• GGUF Q4_M — Smallest GGUF option for constrained hardware
• GGUF Q6_K — Good balance between size and fidelity
• GGUF Q8_0 — Highest-precision GGUF option, with the largest footprint
• FP8 Distilled — Faster generation and rapid iteration
• FP8 Dev — Full model behavior and maximum workflow flexibility
The distilled workflow is the easiest starting point when speed matters. Choose the full FP8 or higher-precision GGUF variants when final quality and flexibility matter more.
INCLUDED RESOURCES
• LTX-2 19B GGUF models: Q4_M, Q6_K and Q8_0
• Standard and distilled FP8 models
• Gemma text encoders
• Video and audio VAE components
• Distilled, detail-enhancement and camera-control LoRAs
• Spatial and temporal upscalers
• Canny ControlNet
• EKKIVOK text-to-video and image-to-video workflows
COMFYUI QUICK SETUP
1. Update ComfyUI before loading an LTX-2 workflow.
2. For native workflows, open Template Library → Video → LTX-2.
3. For GGUF models, install ComfyUI-GGUF and use Unet Loader (GGUF). Place .gguf files in ComfyUI/models/unet/; recent versions can also expose ComfyUI/models/diffusion_models/.
4. Place LoRAs in ComfyUI/models/loras/ and text encoders in ComfyUI/models/text_encoders/.
5. Load the included workflow, install missing custom nodes through ComfyUI Manager, then restart ComfyUI.
6. Select the downloaded filenames in the corresponding loader nodes. A downloaded filename may differ from the name saved inside an older workflow, so relink it once if ComfyUI shows a missing model.
TROUBLESHOOTING
• GGUF model missing: verify that you are using Unet Loader (GGUF), refresh the model list and restart ComfyUI.
• Red or missing nodes: use Install Missing Custom Nodes in ComfyUI Manager, then restart.
• Workflow asks for a differently named file: select the equivalent file manually in the loader node; you do not need to download the entire repository.
• Out of memory: begin with a smaller GGUF quantization or the distilled workflow, reduce resolution and frame count, and avoid loading unused models simultaneously.
• Need help: include your GPU, VRAM, ComfyUI version, workflow name and exact error message in your comment.
CREDITS AND DOCUMENTATION
• Original LTX project: https://huggingface.co/Lightricks
• Official LTX documentation: https://docs.ltx.video/open-source-model/integration-tools/comfy-ui
• Official ComfyUI LTX-2 guide: https://docs.comfy.org/tutorials/video/ltx/ltx-2
• ComfyUI-GGUF: https://github.com/city96/ComfyUI-GGUF
If this hub or one of my workflows helps you, sharing a result and leaving a review is the best way to support continued updates.
Description
This is the Camera control (Dolly Out)
FAQ
Comments (69)
you should add the fp4 dev model. I kept reading don't try it unless you have a 5090, nah dude, runs great on 4090, no speed boost over fp8, but huge memory savings, don't have to quantize or distorch swap, running 1600x960 10 second clip right now, 2 minutes, was taking me 5 minutes on fp8 model because i had to memory swap the model, i can even push more, that was only 19gb usage.
hmmm ....yeah i will upload an FP4 version
doesn anyone know which the most efficient gemma model to use? So many versions are usable now
ive been using the 400mb unsloth version, is it the best? idk, but its 400mb lol
@spiralz if you got a lot of vram and ram i suggest abliaterated version.
if not....use the one with less weight :) that will do the job or use GGUF (i will upload the best one)
@MrReclusive666 basicaly yes it's text encoder :)
@EKKIVOK i still can't get abliatarated working right, i did some videos yesterday, ended up using the 400mb unsloth gemma 3, but tried it with abliaterated, ended up with 2 dudes sifting though rocks? it completely ignored my prompt, but the 400mb gemma 3 gave me exactly what I asked for.
@MrReclusive666 hmm did u use the connector embeddings text encoder with it ?
@EKKIVOK yeah, i did, i think my issue is multigpu, certain things don't like to run if another model it wants to talk to is on another gpu, my assumption is that "connector" needs the model still, but that model is isolated on its own gpu so it can't find it.
@MrReclusive666 i don't use it so i can't help you im really sorry about that but my release of tomorrow (special workflow) will maybe help you out) i will try to edit and include a special version for you :)
@EKKIVOK ill take a look when you release it, i don't look at a lot of workflows on here, because the multigpu thing, don't need to try and cram vae,clip, and model all on one gpu when i have 3 gpu's
@MrReclusive666 oh so the problem comes from there 3 gpu comfyui don't really like that especially since the last updates...
@EKKIVOK its been running fine, just certain things don't like to talk to eachother if they are on separate gpu's, it hasn't been a big deal, just wish i could find a solution to gemma 3, don't like having half a 4090 dedicated to it.
@MrReclusive666 in fact i never had multiple gpu i have only a 5090
@EKKIVOK yeah, i got 2 4090's and a 4080. would like a 5090, but honestly, half tempted to start looking for used 3090's since they had nvlink, get 4 of those at a decent price, bam 96gb vram.
@MrReclusive666 that's really nice and tho i didn't know comfyui can work with multiple card xD i guess i will buy a second 5090 then xD
@EKKIVOK yeah, unfortunately without nvlink, they work independently, so i have 2 24gb cards and a 16 gb card, not a 64gb card, so only comes in handy separating things like vae and clip from unet, but massive unets, still a pain, which is why fp4 working was a nice find, the unet is only 14gb at fp4
is IMG TO video is doing a lot longer than t2v, t2v only like 4minutes 720p, but i2v like 20minutes wth am i doing it wrong,32gb ram/16gbvram
@Seii1 wjhat is exactly your specs ?
@EKKIVOK 4060ti with 16gb vram + 32gb ram, my virtual i sett t0 100gb on nvme
@Seii1 hmm seems a little bit weak for dev version....did you tried distilled version 8 steps ?
if yes what res did u use for your videos ?
@EKKIVOK using q6 720
@Seii1 with 20 or 8 steps ?
@EKKIVOK 8 steps
@Seii1 so try to use it with the distilled lora it should accelerate the process :)
@EKKIVOK btw do use have Workflow for the GGUF version ?
@Seii1 Yeah i made one it will be release tomorrow with some kind of ...surprises :D
hi @Seii1 where did you find the i2v workflow?
@vAnN47 from this post, just download the video and place it into ur comfy
@vAnN47 @Seii1 exactly <3 but the final workflows will be released soon :)
The video was made with a none definitive one, expect better results with the new one !
@EKKIVOK cant wait for the workflow :)
16gb-vram here. I gave up on LTX2 because the second sampler (that one that processes the upscaled latent) runs very slow. I truly believe that in a few weeks or months there will be better models for Us, the people that can not afford an 5090. And yes, the worflows works, I produced around 30 funny videos using it.
si i'm happy that it worked for you too, yes, the spatial upscaler is a little bit slow BUT if you lower the res to 880x440 or 440x880 it will work better and way faster. The maximum for LTX2 is basicaly 720x1280 res. i have a 5090 and able to do 1080p but the result is not better with those classic res :)
@lidianeporto9248 every videos i made in that page are basicaly 720p and 440p :)
so try to lower the resolution upscaler will do a better job and faster ;)
@EKKIVOK Thanks, friend. In fact now I'm very satisfied after I added this to command line "--cache-none --reserve-vram 3 --preview-method none". No OOM (out of memory) 24 fps, perfect lip sync, fast generations. I use gemma q2-gguf.
@lidianeporto9248 You can use my personal Args too and see if it's fit for you :)
--windows-standalone-build --use-sage-attention --fast fp16_accumulation --cuda-malloc --auto-launch
How did you do the One Punch Man video? I have a hard time making anything with anime stuff. What prompted you to look like an anime scene?
hmmmm download the video and drag it on your comfyui workspace :D
Magic can happen .....Just sayin ..... <3 @Agino
@EKKIVOK I did that and didn't work, it can't find the workflow from the video
@Agino so ....wait for it, it come tommorow :)
@EKKIVOK ok i will wait then
@Agino okay <3 trust me you will like those workflows :)
@EKKIVOK ok i will be wait for the good new
Hi bro! How do you think, is there any chance for NSFW LTX 2 like Wan 2.2 in the future?
@Honeyphoria yeah and you know bro ? Check this page this week and you will know why :D
@EKKIVOK OMG. I will support you with all money i can😁 It's gonna be new era of AI NSFW waifu creators!
@Honeyphoria 3 NSFW versions coming soon :)
@EKKIVOK ohhh boy, i'm so excited!! You're a legend!!
Hi! Did I miss something or is there a workflow with the model? :)
sure there is there is 3 :) search ekkivok workflow ltx. but this early stage, i'm working on a mucj much MICH more efficient and advanced workflows for those models :)
I've only got 16 GB VRAM (64 GB DDR4). But:
How do you get these nodes?
VHS_VideoCombine
easy cleanGpuUsed
CM_FloatToInt
Only the 2 first is from your workflow and my comfy is missing those.
Last one is not in your workflow, but in another LTX2 workflow (but then its model is 40 GB...), and btw it's found in "ComfyMath" nodes, which again IS found in Custom Nodes Manager.
The others are not. Not directly at least.
"VHS_VideoCombine" is mentioned in the description of "Video Output Bridge" node. Is that the one to install?
"easy cleanGpuUsed" is not found in that manager. Do I just search it up and (if found) git pull it into custom_nodes?
Also, the name in the "Load Checkpoint" node is different, and apparently I can't click on the name to find another one, it's fixed. I got your "all in one" models, in subfolers of "checkpoints". I also tried having them in "diffusion_models" and "unet". Maybe they should be in another folder?
Also, since I suspect this is needed: How do I load GGUF's? (something about unet nodes but I have no idea which one that shows up in that Nodes Manager).
Yeah I have no idea what I'm doing, I've barely gotten ComfyUI to work (again, this time via yet another install method, but it works the same afaik, I can do some Z-image gens just fine),
Sorry but this is very frustrating.
pretty simple you need to use comfyui manager. Comfyui Manager is able to find and install missing nodes in a workflow :) you NEED that custom node in order to use any workflows properly :)
@EKKIVOK Thanks for replying!
Okay, so maybe there's a simple fix like that, I'll keep trying.
Just FYI or something, but when I open a workflow with missing nodes, I get a pop-up listing missing nodes, Just a list, and I've only tried a simple copy-paste names in Custom Nodes Manager, to no avail ofc because apparently they are packaged with other (custom) nodes.
Sometimes I get a pop up about missing models, but at least they have a download link. (As said, kind of new to ComfyUI).
Sorry for this mess... this is not for your workflow anymore but for the "LTX-2 Text to Video" template (or I guess workflow) that I somehow got in ComfyUI's Templates sidebar button.
I no longer get the list of missing nodes (Unless I use your workflow!), I think all I did was update comfyUI (which I had done earler also).
But I can't select a model. The "Load Checkpoint" node's name field is grayed out. Same for the "LTXV Audio VAE Loader" node. Text encoder node worked (for selecting a model at least).
According to this: https://comfyui.dev/docs/guides/nodes/load-checkpoint/
The checkpoint should be located in the expected place (models/checkpoints). But ofc that doesn't work. (I also tried diffusion_models folder).
Now I do use that extra_model_path.yaml and point to another drive, but not even when copying a checkpoint to ComfyUI's install folder and model/checkpoints/ helped (refreshed browser ofc). I only tried with one smallish checkpoint though. Maybe it's a wrong one that the loader don't recognize, or something else. I have no idea.
I do use subfolders, like "LTX2" for those models, and "Z-Image" for those to try and keep things somewhat organized, as long as I can click and get that drop-down list it works just fine for Z-image, Qwen-TTS and ACE Step 1.5, which are the ones I've made to work in ComfyUI so far.
@wrOngplanet I personally don’t have this issue on my side, but I’m going to push updated workflows as fast as I can today on Civitai and make sure they work as perfectly as possible.
Can you like provide a workflow that DOESNT have renamed models? Wtf is all this shit? Why are there so many downloads. Your CHATGPT copy paste read me is fucking useless. Can you name things like a normal person?
No. and if you're not ok with that there is other releases.
Good luck. youy didn't made the job to unite all that different kind of models in one post.
I did it, stop complaining like a baby. be thankfull for someone did that for you.
Have a good night <3
@EKKIVOK imagine telling your customer stop complaining when you can't even do a simple job LOL.
@PixivTdog
Do the d*mn job first, then we can talk.
Stop whining about free stuff that people spent their own time uploading to a site.
Dude, it’s FREE, so quit complaining or go make it yourself.
If you don’t know how to do it, stop being rude and just ask nicely for once.
Don’t come crying when the answer matches the vibe of your first message.
I’m not here to deal with people like that.
When customers pay me, I do the work, I update it however they want, done.
Here? It’s free. All the useful links and direct download links are right there in front of your eyes and you’re still crying like a baby.
Come on, man… move on.
Oh and another thing it literally says “Upload in progress”.
The post isn’t even 100% finished. I’ve got a family, a job, hobbies…
I’m not sitting here 24/7 waiting for your personal satisfaction, especially not when someone’s bitching like that over free content.
Not happy? The original Hugging Face repo is linked in the description too.
Deal with it. But stop raging at free upload. Or DIY.
Final message i'm done talking to ppl like that.
Get a job
@TononMauricio "The post isn’t even 100% finished. I’ve got a family, a job, hobbies…" learn how to read bro.
@EKKIVOK it was quite rude of you to release your workflow for free, I guess ... or whatever they are complaining about. just joking of course, simply wanted to send you some good vibes. don't let them get to you and keep your spirits high. best wishes. ;)
@philcore thanks bro, all my releases are free, it's for that reason when i receive that kind of comments it make me mad....xD it's the rules i guess....but not the good ones...
@EKKIVOK I didn't read anything you said, I guess you enjoy using twitter.
@EKKIVOK there's always a troll or two - just ignore
@EKKIVOK I was obviously talking to the comment creator. such a fragile ego gez
Hello,
I am trying to use the workflow that you posted "EKKIVOK - LTX2 IMG2IMG.json", however I can not seem to find "proj_linear.safetensors" anywhere, might you know how I can get it? Here is the error message:
"LTXVGemmaCLIPModelLoader
No such file or directory: C:\Users\User1\Downloads\ComfyUI\models\tts\Qwen3-TTS\Qwen3-TTS-12Hz-1.7B-VoiceDesign\proj_linear.safetensors"
can you have a gguf workflow? i can't seem to redo this in gguf models
your gemma 3 is corrupted, also you didnt provide any explanation of how to use this fp8 ver (I'm assuming)? The workflow u have says download entire gemma3 repo
Details
Files
ltx219BCompleteToolkitGGUF_ltx2CameraDOLLYOUT.safetensors
Mirrors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
ltx-2-19b-lora-camera-control-dolly-out.safetensors
VG2-19b-lora-camera-control-dolly-out.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.