SMOOTH WORKFLOW LTX
LTX Workflows are some crazy spagetti aren't they? With this workflow I tried to simplify the UI to the basics while keeping the same UI setup as the Smooth Workflow v4.0 AIO for WAN 2.2.
All images used on the Showcase use the same set of loras, weights and configurations.The only difference are the images used for I2V and Positive Prompts for the Img2Vid and Txt2Vid generations.
So, before we begin, all videos on showcase used the following:
Model -> ltx-2.3-22b-dev_transformer_only_fp8_scaled
Lora -> ltx-2.3-22b-distilled-lora-dynamic_fro09_avg_rank_105_bf16
Text Encoder -> gemma_3_12B_it_fp8_scaled
Text Encoder -> ltx-2.3_text_projection_bf16
VAE -> vae/LTX23_audio_vae_bf16
VAE -> vae/LTX23_video_vae_bf16
VAE -> taeltx2_3
Upscaler -> ltx-2.3-spatial-upscaler-x2-1.1 (update)
Lora-> SmoothMix Animation LTX - v1.0
Of course you should use the setup of models/loras/encoders that your hardware can handle and your own preferences. ^^
After downloading all thats necessary make sure to these extensions installed and updated:
ComfyUI-GGUF (In case you use a GGUF model)
Inital Setup
Choose your workflow: Image2Video or Text2Video.
In the middle of the workflow you set all the Models and Loras your generation will use.
If you are going to use a GGUF Model remember to connect the GGUF Node to the Lora Node.
You can make very specific Positive Prompts when using LTX. Check the showcase for examples. The default Negative Prompt should be enough for most cases - if necessary delete or add prompts to it.
Set how long you want your video to be. Its set to 10 seconds by default.
IMG2VID and TXT2VID WORKFLOW
Both workflows follow the same format. Set the resolution and the frame rates for the Audio and Video. Its set to 25 frames by default as thats the LTX default setting. Some people also try setting it to 30 frames with good results, so give it a try as well.
There isn't much else to try since any small change can make a huge difference: The Seed/Noise, the Sampler (euler is set by default) and CFG are readily available for new values. So change any of them if needed.
But for mode advanced users here is how everything else its organized:
There are four Subgraphs: Image Setup, Sampler 1, Sampler 2 and Decode. Each has a very simple, easy to follow, set of nodes. It should make it easy to find or change what you think its necessary.
Image Setup has a very simple upscaling. Its like this to make generation fast. Make changes to upscaling methods here.
Sampler 1 and Sampler 2 have the samplers where you can change the ManualSigmas.The workflows have different default ManualSigmas on their Sampler 1. Both workflows have the same default ManualSigmas on their Sampler 2. Make sure to change it to your preferred values.
Decode has the final decoder. It uses all the standard decoders. Very Simple. If use have a prefered VAE decoder just change it here.
Description
FAQ
Comments (17)
I think I may have found my new favorite LTX2.3 workflow!
A little unsure how the Adaptive prompts work but I don't think that I'll be much of an issue.
Looking forward to improvements and future releases!
Also in case you wanted to know, there was an update to the upscaler, that fixed some bugs at the end of videos.
https://huggingface.co/Lightricks/LTX-2.3/blob/main/ltx-2.3-spatial-upscaler-x2-1.1.safetensors
Anyone has successfully used Google new Turboquant algorithm to make 30+ seconds videos on 16gb of vram? Cause that should be possible.
I really like how this workflow is structured. Sadly, it gets stuck at the negative conditioning (red block). Nothing happens, nothing in the logs. Comfy is up to date. 5090 + 64GB, should normally be fine. Any idea?
EDIT- Okay, so I WASN'T using everything from the workflow as instructed, derp. The issue was I wasn't using the recommended "ltx-2.3-22b-distilled-lora-dynamic_fro09_avg_rank_105_bf16" lora, I didn't realize it was mandatory cause I'm a doofus.
Hello! Thank you very much for all your hard work- I still use your WAN workflow all the time, it's very friendly for a comfytardlet like myself, haha
That being said, I'm completely new to LTX and dipping my feet in now that the scene has matured a bit- but for some reason I get very, very poor results with your LTX workflow resembling blurred fever dreams complete with audio distortion. I'm just using all the default settings and using this as the checkpoint-
https://civitai.com/models/2447875/ltx23-10eros?modelVersionId=2752410
This happens with both I2V and T2V.
I asked LLMs and they gave me garbage advice even after giving them the json from this workflow, could some kind soul help a lowly script kiddie tardlet like myself out? :<
What am I doing wrong? :(
Newer to LTX2.3 myself and just trying to use their standard setup but getting blurry results to.
@nostela You are likely smarter than I, haha, but for me it was because I wasn't using that distillation lora. LTX results are definitely kinda all over the place I see, but my fever dream messes are at least gone once I started using the lora.
@Straitjacket oh thanks! I will test this and see. I managed to find another workflow that worked well for me. I appreciate the help! Haha
Where i need to add this upscaler model? i added it to folder upscale_models and don't work. i even added it to every folder and still can't find it in loader.
Solved, i need to create new folder "latent_upscale_models" manually, but no one told about it.
Next problem with information about downloading custom nodes - if i download manually, unpack it i need to extract folder below to up, then i need to change folder name removing -main because comfyui doesn't see added nodes.
When u try to generate video with your workflow teh soudn is jsut some kidn of errors adn the video is blur.What i do wrong?or it msut be liek specific size of image for video?
i keep getting subtitles. I have it in the negative prompt box... how to get rid of them?
Супер рабочий процесс ! благодарю за труды вы как всегда на высоте!
This LTX workflow works well; I just changed two nodes and replaced them with other compatible models. Thanks for sharing
@DigitalPastel loving this, great work!
any chance for this workflow to be updated with ltx director in the near future? o.O
tyvm for your great work!!!
VAELoader KJ failed
This node threw an error during execution. Check its inputs or try a different configuration.