Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

LTX-2.5 is native in ComfyUI, day zero with the weights Model highlights: → Diffusion Fidelity Rendering - compute allocated by scene complexity, not spread evenly → Diffusion Video Decoder - replaces VAE decoding, sharper faces and legible text → Native multi-shot - one run, multiple connected shots, consistent across...

2,406,572 Aufrufe • vor 3 Monaten •via X (Twitter)

17 Kommentare

Profilbild von ComfyUI
ComfyUIvor 3 Monaten

Click the link below:

Profilbild von Prince does AI
Prince does AIvor 3 Monaten

legible text in video is the part i care about

Profilbild von ComfyUI
ComfyUIvor 3 Monaten

That I 100% agree with.

Profilbild von Rob Hoffman (No longer redacted)
Rob Hoffman (No longer redacted)vor 3 Monaten

In my meager tests, it's a big speed improvement. Looking forward to giving it a real workout this weekend.

Profilbild von LabMike 3D
LabMike 3Dvor 3 Monaten

I'm currently testing LTX 2.5 on my RTX 3060 12GB, and it's pretty fast with great quality. Cheers!

Profilbild von 泰多宝
泰多宝vor 3 Monaten

The decoder swap is the one that changes what you can ship. A VAE trained on reconstruction loss averages high-frequency detail, and faces and glyphs are made of exactly that — you get texture that looks like text rather than text. A diffusion decoder generates the detail instead of averaging it, which is why it comes out legible. The tradeoff rides along though: legible isn't correct, since it synthesises plausible strokes rather than recovering the encoded ones.

Profilbild von LabMike 3D
LabMike 3Dvor 3 Monaten

LTX 2.5 on RTX 3060: Impressive Speed & Quality Out of the Box

Profilbild von JSFILMZ
JSFILMZvor 3 Monaten

10 sec max duration in 2026 is wild

Profilbild von smallzero
smallzerovor 3 Monaten

I got you! Already created a video showing how to launch inside ComfyUI

Profilbild von Vini
Vinivor 3 Monaten

Allocating compute by scene complexity instead of evenly is smart engineering.

Profilbild von Mr Zoulll
Mr Zoulllvor 3 Monaten

Mon RTX 5090 l'adore. 🥰

Profilbild von Hao
Haovor 3 Monaten

拿中文语境做了2次实验,跟h3对比…🤦

Profilbild von Nazar
Nazarvor 3 Monaten

Replacing VAE decoding is the line people will skim past and shouldn't. Faces and text going soft in the decode has been the tell on every open video model. If that is genuinely fixed, it changes what these are usable for beyond b-roll.

Profilbild von Narine Saw
Narine Sawvor 2 Monaten

Ltx lol Mini Max h3 !!!

Profilbild von derrick has started yet another project
derrick has started yet another projectvor 3 Monaten

No more Israeli models thanks

Profilbild von ReAiLity Labs
ReAiLity Labsvor 3 Monaten

Is a new template being pushed out in Comfyui or did the existing ltx templates get updated?

Profilbild von Edgex
Edgexvor 3 Monaten

Wow, the Diffusion Fidelity Rendering really changes the game - love how it adapts to scene complexity.

Ähnliche Videos

Auto regressive LLMs are officially on notice. run Gemma 4 26B diffusion gguf with llama.cpp Google just dropped DiffusionGemma-26B, and it completely flips how we generate text. instead of predicting words one by one, it generates 256 tokens in parallel using bi-directional attention. its like stable diffusion, but for language. the model starts with random text "noise" and iteratively refines and self-corrects the entire block in real-time to fix formatting and reasoning errors on the fly. since it’s a Mixture of Experts (MoE) that only activates 3.8B parameters during inference, it fits perfectly on consumer hardware. You can run the Q4_K_M quant with an 18GB VRAM budget on a single RTX 3090 or RTX 4090 with exceptional throughput. Tested on Ubuntu 22 with CUDA 13.1 using the cutting edge experimental llama.cpp branch. Here is how to compile and run it with the live terminal denoising visualizer: # 1. Clone & check out the experimental PR (#24423) - 1) git clone && cd llama.cpp -git fetch origin 2) pull/24423/head:diffusiongemma && --git checkout diffusiongemma # 2. Build with CUDA support 1) cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native 2) cmake --build build -j $(nproc) --config Release --target llama-diffusion-cli # 3. Run with live visual denoising (llama.cpp flags) ./build/bin/llama-diffusion-cli \ -m /path/to/diffusiongemma-26B-A4B-it-Q4_K_M.gguf \ -ngl 99 -cnv -n 2048 --diffusion-visual Watch the video below to see the live --diffusion-visual canvas iteratively de noising the prompt output in real time. guide and unsloth's hugging face GGUF model links are in the comments below! Is auto regressive generation officially legacy tech? Let me know what you think.

Alok

52,656 Aufrufe • vor 5 Monaten