NEW LTX-2 Workflows here: https://civarchive.com/models/2318870
Workflow: Image -> Autocaption (Prompt) -> LTX Image to Video
LTX Prompt Enhancer (LTXPE) might have issues with latest Comfy and Lightricks update
Update July 20th 2025: GGUF Models for LTX 0.9.8:
Distilled model, works with V9.5: https://huggingface.co/QuantStack/LTXV-13B-0.9.8-distilled-GGUF/tree/main
Dev model, works with V9.0: https://huggingface.co/QuantStack/LTXV-13B-0.9.8-dev-GGUF/tree/main
(see "Model Card" in above links for LTX 0.9.8 VAE and textencoder downloads)
V9.5: LTX 0.9.7 Distilled Workflow supporting LTX 0.9.7 Distilled GGUF Model.
There is a workflow with Florence and another one with LTX Prompt Enhancer (LTXPE)
GGUF Model can be downloaded here:
https://huggingface.co/wsbagnsv1/ltxv-13b-0.9.7-distilled-GGUF/tree/main
VAE and Textencoder are identical to previous LTX 0.9.6 model (see V8.0 below)
LTX 0.9.7 Distilled is using only 8 steps and is very fast.
V9.0: LTX 0.9.7 Workflow supporting LTX 0.9.7 GGUF Model.
There is a workflow with Florence and another one with LTX Prompt Enhancer (LTXPE)
GGUF Model can be downloaded here:
https://huggingface.co/wsbagnsv1/ltxv-13b-0.9.7-dev-GGUF/tree/main
VAE and Textencoder are identical to previous LTX 0.9.6 model (see V8.0 below)
LTX 0.9.7 is a 13billion parameter model, previous versions only had 2b parameters, therefore it is more heavy on Vram usage and requires longer process time. Try V8.0 below with model 0.9.6 or V9.5 for very fast rendering.
V8.0: LTX 0.9.6 Workflow (dev and distilled GGUF model in same workflow)
there is a version with Florence2 Caption and a version with LTX Prompt Enhancer (LTXPE)
GGUF Models (Dev & Distilled) can be downloaded here:
https://huggingface.co/calcuis/ltxv0.9.6-gguf/tree/main
vae: pig_video_enhanced_vae_fp32-f16.gguf
Textencoder: t5xxl_fp32-q4_0.gguf
V7.0: LTX 0.9.5 Model Version GGUF with Wavespeed/Teacache.
LTX 0.9.5 GGUF Model and VAE: https://huggingface.co/calcuis/ltxv-gguf/tree/main
(vae_ltxv0.9.5_fp8_e4m3fn.safetensors)
Clip Textencoder: https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main
There are 2 worklfows, a main workflow with florence caption only and additional one with florence and LTX prompt enhancer. Setup with Wavespeed (bypassed by default, Strg+B to activate)
workflow works with all GGUF models: 0.9 / 0.9.1 / 0.9.5
uncensored LLM for Prompt enhancer: https://huggingface.co/skshmjn/unsloth_llama-3.2-3B-instruct-uncenssored
-Outdated (march 2025)- V6.0: GGUF/TiledVAE Version & Masked Motion Blur Version
Updated the workflow with GGUF Models, which save Vram and run faster.
There is a Standard Version, which uses just the GGUF Models and a GGUF+TiledVae+Clear Vram Version, that reduces Vram requirements even further. Tested the larger GGUF model (Q8) with resolution of 1024, 161 frames and 32 steps , the GGUF Version peaked Vram usage at 14gb, while the TiledVae+ClearVram Version peaked at 7gb. Smaller GGUF Models might reduce requirements further.
GGUF Model, VAE and Textencoder can be downloaded here:
(Model&VAE): https://huggingface.co/calcuis/ltxv-gguf/tree/main
(anti Checkerboard Vae): https://huggingface.co/spacepxl/ltx-video-0.9-vae-finetune/tree/main
(Clip Textencoder): https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main
You can go for the GGUF Version with 16gb+ and the TiledVae+ClearVram with less than 16gb Vram.
Masked Motion Blur Version: Since LTX is prone to motion blur, added an extra group to the workflow which allows to set a mask on input image, apply motion blur to mask, to trigger specific motion. (sounds better than it actually works, useful tho in some cases). GGUF and GGUF+TiledVAE+ClearVram version included.
V5.0: Support for new LTX Model 0.9.1.
included an additional workflow for LowVram (Clears Vram before VAE)
added a workflow to compare LTX Model 0.9.1 vs LTX Model 0.9
(V4 did not work with 0.9.1 when the model was released (hence v5 was created), this has changed as comfy & nodes were updated in the meantime, now you can use both Models (0.9 & 0.9.1) with V4, also with V5. Both have different custom nodes to manage the model, other than that, both versions are the same. If you run into memory issues/long process time, see tips at the end)
-Outdated (march 2025)- V4.0: Introducing Video/Clip extension :
Extend a clip based on last frame from previous clip. You can extend a clip about 2-3 times before quality starts to degenerate, see more details in the notes of the worflow.
Added a feature to use your own prompt and bypass florence caption.
V3.0: Introducing STG (Spatiotemporal Skip Guidance for Enhanced Video Diffusion Sampling).
Included a SIMPLE and an ENHANCED workflow. Enhanced Version has additional features to upscale the Input Image, that can help in some cases. Recommend to use the SIMPLE Version.
replaced the height/width Node with a "Dimension" node that drives the Videosize (default = 768. increase to 1024 will improve resolution, but might reduce motion, also uses more VRAM and time). Unlike previous Versions, Image will not be cropped.
Included new node "LTX Apply Perturbed Attention" representing the STG settings (for more details on values/limits see the note within the workflow) .
Enhanced Version has an additional switch to upscale Input Image (true) or not (false). Plus a scale value (use 1 or 2) to define the size of the image before being injected, which can work a bit like supersampling. As said, not required in most cases.
Pro Tip: Beside using the CRF value at around 24 to drive movement, increase the frame rate in the yellow Video Combine node from 1 to 4+ to trigger further motion when outcome is too static.
Node "Modify LTX Model" will change the model within a session, if you switch to another worklfow, make sure to hit "Free model and node cache" in comfyui to avoid interferences. If you bypass this node (strg-B) , you can do Text2Video.
V2.0 ComfyUI Workflow for Image-to-Video with Florence2 Autocaption (v2.0)
This updated workflow integrates Florence2 for autocaptioning, replacing BLIP from version 1.0, and includes improved controls for tailoring prompts towards video-specific outputs.
New Features in v2.0
Florence2 Node Integration
Caption Customization
A new text node allows replacing terms like "photo" or "image" in captions with "video" to align prompts more closely with video generation.
V1.0: Enhanced Motion with Compression
To mitigate "no-motion" artifacts in the LTX Video model:
Pass input images through FFmpeg using H.264 compression with a CRF of 20–30.
This step introduces subtle artifacts, helping the model latch onto the input as video-like content.
CRF values can be adjusted in the yellow "Video Combine" node (lower-left GUI).
Higher values (25–30) increase motion effects; lower values (~20) retain more visual fidelity.
Autocaption Enhancement
Text nodes for Pre-Text and After-Text allow manual additions to captions.
Use these to describe desired effects, such as camera movements.
Adjustable Input Settings
Width/Height & Scale: Define image resolution for the sampler (e.g., 768×512). A scale factor of 2 enables supersampling for higher-quality outputs. Use a scale value of 1 or 2. (changed to dimension node in V3)
Pro Tips
Motion Optimization: If outputs feel static, incrementally increase the CRF & frame rate value or adjust Pre-/After-Text nodes to emphasize motion-related prompts.
Fine-Tuning Captions: Experiment with Florence2’s caption detail levels for nuanced video prompts.
If you run into memory issues (OOM or extreme process time) try the following:
use the LowVram version of V5
use a GGUF Version
press "free model and node cache" in comfyui
set starting arguments for comfyui to --lowvram --disable-smart-memory
see the file in your comfyui folder: "run_nvidia_gpu.bat" edit the line: python.exe -s ComfyUI\main.py --lowvram --disable-smart-memory
switch off hardware acceleration in your browser
Credits go to Lightricks for their incredible model and nodes:
Description
Introducing Video/Clip extension for LTX/STG Image to Video
(outdated since march 2025)
FAQ
Comments (45)
Great workflow! Is it possible to change the output style? For example, take an image and generate a video that looks like something out of an anime, video game, etc?
First of all: the results are outstanding. But I'm a long way from a minute! A workflow has currently been running for about 15 minutes, and it says it will take another 30 minutes. What am I doing wrong? I have adopted all the settings as in the example. I have 16 GB VRAM and 64 GB RAM.
Guess you ran into shared memory for some reason, which can make render time grow like 10x. You could try "free model and node cache" in comfy and try again. I am on 16gb Vram/64gb Ram as well.
@tremolo28 I was sure, I did it before... I tried it again and it was 55 and 79 seconds. Thank you very much!
I had no problem getting the provided cat image to animate, but the video is just static with any of my own images. I've bumped up the salt and pepper as recommended.
Make sure, the extend clip switch is not set to "True", otherwise it will bypass both parameters to trigger motion. If you switch it to "False" it should work.
could you up for 0.91 model please? When i choose it with v4.0 of your latest workflow it gives me a huge error, it needs a VAE not, but i have no clue where to add it
Seems to require a custom vae node. Will have a look.
thank you!
@tremolo28 Highly appreciated, thanks for all your work 🎖️🎖️🎖️
@tremolo28 Commit 418eb70 Support new LTXV VAE.
got it running with 0.91, need to run some testing, workflow is released as experimental workflow for testing.
it tells me, im missing LTX Video loader, LTXApplySTG, LTXV Model Configurator, STG Guider, LTXVshiftSigmas, when i check install missing, it finds nothing, do i have to install them via url, github? v4.0 workflow works fine
@NeuroFunkeR Just install ComfyUI-LTXVideo nodes
@EliteLensCraft yeah for some reason i can't getting error after restart, that it was unable to install, LTX lighttricks is installing just fine
@NeuroFunkeR What also helped is updating all nodes to the current version, maybe some dependencies thing
The experimental workflow essentially is working @tremolo28 , however I'm running in the same issue like described here: https://www.reddit.com/r/StableDiffusion/comments/1hhz17h/comment/m2v4hi0 - VAE Decode is throwing random OOM error, even with unloading models...
@EliteLensCraft ill reinstall comfy, it probably messed up something after updating
now it finally seems to work
@NeuroFunkeR Did you reinstall the standard portable version of comfy and then not update it? I've been getting those OOMs with the newest update.
@dfree3305149 yes i did and now i do get the same OOM errors on 16gb VRAM, i tried to enable shared memory in control panel, doesn't seem to help
Hey great job! It's a shame that with 0.9.1 it requires more vram for each iteration (it no longer fits in 8gb vram so it becomes much slower). Also the last step (vae decoder) seems to consume more vram and also makes it use shared vram on 8gb cards.
yes, v0.91 seems to use more vram. Did you try to place a "Clean Vram" node b4 sampler and b4 vae? The node is avail. as a custom node in the easy-use pack. Pls let me know if that works in case you try it.
@tremolo28 Thanks for this tip, I didn't know this node existed... I plugged it right before the VAEdecode and now it runs quite nice at higher resolutions without OOM...!
@zampano what node do i need for that? When i search for clean vram nothing i find
@tremolo28 I haven't tried it, but it should work (as others commented) for that last step. The bottleneck, however, is in the actual video generation, which also requires more vram.
@NeuroFunkeR You can find the "Clean Vram used" node in the ComfyUI-Easy-Use package in Github. After installing, restart comfy and refresh the browser, then right click and add the node (it's in the Easy Use package -> Logic tab)
@perros19 thanks i already found LatenGC which does the same basicaly
@NeuroFunkeR https://github.com/yolain/ComfyUI-Easy-Use
@dfree3305149 yeah thanks ive got it too already, but it doesn't help. I did put it before VAE and before Adanced samperl, still getting OOMs
omg this is fireeee
https://www.youtube.com/watch?v=dmZGNzGj4Ps
Created music video using LTX v091
It doesn't say I'm missing any nodes but I get this error.
VideoCombine.combine_video() got an unexpected keyword argument 'crf'
24 salt, 4 pepper.
same for me. this is why i avoid these overly complicated workflows. takes hours to get them working. faster to just build from scratch.
Update the VideoHelperSuite custom nodes - fixed it for me
Did you get a fix? Updating videohelpersuite didn't work for me, same issue
@Spurdo I've given up for now. It just kept crashing.
What are your thoughts on the new Model LTX 0.91 so far? How does it work for you compared to LTX 0.9 from V4?
My observations so far on LTX 0.91:
- a bit more trigger happy for motion
- animation quality not sure if it is improved
- faster on it/sec. but VAE can take longer
- higher Vram usage
- Text overlays, like unreadable credits or subtitles.
- Made T2V comparison and 0.9 wins almost all the time over 0.91
i haven't compared much yet, but i think the motions are better with 0.91
have saved an additional workflow to experimental, that allows to compare result of 0.91 vs 0.9 model in 1 sheet.
I feels like 0.91 overall runs faster then 0.90, I am only using img2video, I am not sure its configuration issue
I am struggling to get 10sec videos in under a minute. I am using 1540x1540 images as reference so I am guessing these dimensions are not the baseline. That being said, I am new to ComfyUI as well as modifying workflows. Where can I go to edit the output image dimensions, I'm not certain which node. Again, BRAND NEW. You legit inspired me today to check this out <3
Also, after queuing an image for img2vid, I notice that my GPU (4090) does not kick in right away. Is this some pre-processing taking place? I'm not impatient, just trying to understand this as far as background processes go.
