CivArchive
    Lightricks LTXV 2.3 - Dev

    This model card focuses on the LTX-2.3 model, which is a significant update to the LTX-2 model with improved audio and visual quality as well as enhanced prompt adherence. LTX-2 was presented in the paper LTX-2: Efficient Joint Audio-Visual Foundation Model.

    💻💻 If you want to dive in right to the code - it is available here. 💾💾

    LTX-2.3 is a DiT-based audio-video foundation model designed to generate synchronized video and audio within a single model. It brings together the core building blocks of modern video generation, with open weights and a focus on practical, local execution.

    LTX-2 Open Source

    Run locally

    Direct use license

    You can use the models - full, distilled, upscalers and any derivatives of the models - for purposes under the license.

    ComfyUI

    We recommend you use the built-in LTXVideo nodes that can be found in the ComfyUI Manager. For manual installation information, please refer to our documentation site.

    Description

    FAQ

    Comments (34)

    joausyq12Mar 7, 2026· 7 reactions
    CivitAI
    It looks like a fantastic model but it's too much for my potato computer 😭😭😭

    etherlothMar 9, 2026· 1 reaction

    Look for unsloth GGUFs in huggingface, you can run it in a GPU with as lil as 8 GB VRAM.

    elevendrMar 9, 2026

    @etherloth Me with 6 GB VRAM

    jaiswalpranav95296Mar 29, 2026

    95% people use cloud, i will teach you how i get 105$ credit every month without referal link.

    Spitfire_502Apr 7, 2026

    @etherloth I tried looking. Can you give me a link?

    Raw_FantasyMar 7, 2026· 5 reactions
    CivitAI

    Dear Lightricks team, thank you for the wonderful free model. My joy is mixed with a lack of understanding why it's free, but I'm taking it :)

    aomyouza2543Mar 9, 2026· 1 reaction
    CivitAI

    @AiMetatron @DigitalPastel @taek75799 @darksidewalker @FX_FeiHou

    BoneMechMar 21, 2026· 2 reactions
    CivitAI

    Great model and incredible to not see this locked behind a paywall. Tons of respect!

    InsanE_GeNMar 25, 2026· 6 reactions
    CivitAI

    А на какой машине ее запустить то можно?

    pork69Apr 4, 2026· 8 reactions
    CivitAI

    jesus fuck, i choked on my drink when i saw that its fucking 42 gigs

    elevendrApr 18, 2026

    You can get quantized models that make the model smaller so it can run on your hardware like GGUF.

    jayjay881Apr 6, 2026· 7 reactions
    CivitAI

    How can people run this? 50GB VRAM?

    TheP3NGU1NApr 9, 2026· 1 reaction

    gpu rentals if you want the full thing. otherwise distilled and quantified version of it are now out. still need a decently large amount of vram tho.

    yuduz367Apr 12, 2026· 1 reaction

    dynamic vram in comfyUI

    watsont474Apr 12, 2026

    Professional GPUs. Blackwell RTX 5000 with 48 or 72 GB of VRAM. Blackwell RTX 6000 with 96 GB of VRAM. These GPUs can be bought from places like Central Computers, but the cost is $5000-$9000. The minimum LTX recommends to have it run at all is 32 GB of VRAM, which is what the 5090 has.

    Valex_198318May 5, 2026· 4 reactions

    I have 12GB VRAM and works just perfect!

    jayjay881May 17, 2026

    @Valex_198318 How do you do that?

    SlimeGirlsFTWMay 25, 2026· 2 reactions

    @jayjay881 For starters, just open templates in Comfyui, look for the LTX 3.2 I2V, T2V, or both, and open them. Then just install all that is missing. Hint: both Dev and Distilled will be missing. Honestly, once you have everything installed form the "errors", throw a little prompt in at 512x512, 24FPS(25, actually), 5 seconds(6, actually), put the ckpt, Lora, Encoder, and upscale model in that you downloaded, hit run. Easy peasy. Just don't go too high on res (H x W) and length/duration, and you'll be good. Also FPS =~24-50. Again, depending on your system. Don't overdo it. I would slowly up duration if you want to try it, too big of a jump and you'll crash the program and possibly your PC. I stay around 30(31, actually) seconds and usually 512x512, but I did 12... 1268? 1228? I don't remember, and of course 1024 x 1024.

    attackofthepancakesJun 20, 2026

    @jayjay881 get a lot of system memory and time.

    bassambarrakApr 13, 2026· 3 reactions
    CivitAI

    I have the distilled model, and it gets the job done, while it gets the PC hanging, 32GB RAM and RTX3080 10GB, (using Pinokio)

    faksfkwaf22Apr 14, 2026

    Same here. Have you managed to get the upscalers work? Mine crashes whenever during gen when I activate them.

    JamesMccay1995AiApr 28, 2026

    how long does it take you to render out segments?

    12175452May 14, 2026· 2 reactions
    CivitAI

    Tried both, the funny thing is dev version disfigures faces beyond recognition, while the distilled holds its consistency much better

    Pete0Jun 3, 2026· 5 reactions
    CivitAI

    anyone else having issues generating on site? I repeatedly have failed generations, almost 2 weeks now.

    12276704Jun 4, 2026· 1 reaction

    Same here. I did get one video to generate today, and a few last week, but with addons it seems to not work. Both dev and distilled.

    Pete0Jun 5, 2026· 1 reaction

    @dphalereus954 I tried on 2 browsers and it seems to have less problems on chrome. I also noticed that for whatever reason as soon as I removed the NSFW content from the prompt I got 2 consecutive successes. Could be just a fluke though but it'd been a prompt id been trying for a few days.

    EbenezerDanglewoodJun 7, 2026· 1 reaction

    Success rate went from ~60% to 20% to 0% over the past 2 weeks or so. Every single job has been failing for the past 3 days.

    RackedStacks818Jun 10, 2026

    @EbenezerDanglewood I mean, has anyone gotten any video generation to work in the past week?

    trestonwinkleJun 15, 2026

    Yeah, it's like 20% success rate. Longer clips have a greater chance of failing it seems

    Pete0Jun 15, 2026

    @RackedStacks818 haven't gotten one to work at all I periodically try and times I don't think are high traffic and still nothing. The error seems to be faster tough haha

    Pete0Jun 15, 2026

    @trestonwinkle you've had actual success recently?!

    trestonwinkleJun 15, 2026

    @Pete0 Yeah, just by brute forcing it by attempting again and again.
    https://civitai.red/posts/29200580
    https://civitai.red/posts/29198001

    Both clips are around 8 seconds. It seems to me longer clips have a greater chance of failing.

    LiterlynJul 3, 2026
    CivitAI

    Good

    RoyalCutiesJul 23, 2026
    CivitAI

    Great Model. Works awesome with loras

    Checkpoint
    LTXV 2.3
    by ltxvideo
    Download (Beta) View on CivitAI
    base model

    Details

    Downloads
    1,548
    Platform
    CivitAI
    Platform Status
    Available
    Created
    3/11/2026
    Updated
    8/4/2026
    Deleted
    -

    Files

    lightricksLTXV23_dev.safetensors

    Size:
    42.98 GB
    SHA256:
    7ab7225325bc403448ea84b6db2269811a880e5118cd2ee2b6282a93d585016f

    Mirrors

    HuggingFace (119 mirrors)
    CivitAI (2 mirrors)
    ModelScope CN (1 mirrors)

    lightricksLTXV23_dev_trainingData.zip

    Size:
    21.63 MB
    SHA256:
    c74f14b8150d086064279d98a6c7774dc45819bf6b084ee23feaa3ded8bc5764

    Mirrors

    CivitAI (2 mirrors)