VERSION 2
Adds prompt relay for better control and adherence, also adds video frame interpolation for smoother frame rates during generation.
"film_net_fp16.safetensors" goes in:
Models/frame_interpolation
prompt relay custom node git clone link:
git clone https://github.com/kijai/ComfyUI-PromptRelay
(Clone to custom_nodes folder)
VERSION 1
base workflow for ltx 2.3 distilled first last frame
FILES:
OPTIONAL Kijais fp8 Scaled (requires load diffusion model node instead of unet loader node and replaces the gguf entirely. )
https://huggingface.co/Kijai/LTX2.3_comfy/tree/main/diffusion_models
Distilled ggufs
https://huggingface.co/unsloth/LTX-2.3-GGUF/tree/main/distilled
Gemma 3_12B FP4 text encoder
Audio VAE
https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/vae/LTX23_audio_vae_bf16.safetensors
Video VAE
https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/vae/LTX23_video_vae_bf16.safetensors
Text Projection text encoder
https://huggingface.co/Kijai/LTX2.3_comfy/tree/main/text_encoders
Abliterated NSFW LoRA
https://huggingface.co/Comfy-Org/ltx-2/tree/main/split_files/loras
Description
gguf for first last frame
FAQ
Comments (9)
Good workflow. Clean. thx.
Hi. Is latent upscale process included? I cannot find the part
You can string in the latent upscale method but its not used for this workflow technically, it uses the spatial upscaler
@realrebelai i don't see the spatial upscaler node though. i wanted to confirm if the model is properly loaded but cannot find the model loader.
@willshawn2519 its in there forsure, i cant remember off hand exactly where in the workflow because im not by my pc at the moment but i know its in there 🫶 keep checking
I also don't see the upscaler node
@MiddlingMaker yea so i misunderstood what workflow we were talking about. this workflow doesnt call for upscaling but you could technically just string in the original workflows uspcaling section and chain it to the latents and model. its fairly simple just time consuming because of how the workflows structured. if you struggle with it, let me know and i can probably just string it together at some point. working on as few other things right now
@realrebelai tbh i prefer current workflow without upscaler after several tries. other workflows usually resize the image into half then spatial upscale to save vram usage, but it destroys the details of the image.
Works quite well, thanks for sharing. I've been experimenting with LTX FLF for a few days. There are more ways to go about it than with using WAN, of course I started with the complicated way instead of dead simple guidance like this. I'm going to see how Florence does for beefing up the conditioning a bit, but aside from conditioning stuff, the quality of the image is great as it is. Using the abliterated text encoder and fp8 distillation, no gguf. Nice work.