stay tuned 😎
The company behind FLUX.1 announced a new video model.
One thing that stands out is the consistency of the character, clothes and style between the cuts at the beginning of the video.
This looks extremely promising.
00:00
Andreas Blattmann
371 posts
- Excited to get SDXL-Turbo out 🚀by combining the some of the greatest ingredients in generative modeling - diffusion, score-distillation and adversarial training - this model achieves high-quality samples in only 1 step code: github.com/Stability-AI/g… paper: stability.ai/research/adver…
- Happy to be finally able to share our text2vid project for #CVPR2023 with the amazing @robrombach @HuanLing6 @timudk @seungkim0123 @FidlerSanja and @karsten_kreis. Unfortunately we cannot release the weights. That's why I joined @StabilityAI to work on OS video models 🚀Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Models turn the publicly available, state-of-the-art text-to-image LDM Stable Diffusion into an efficient and expressive text-to-video model with resolution up to 1280 x 2048 abs: arxiv.org/abs/2304.08818
00:00 - Super excited to share the video model we've cooked up in the past few months! We show a lot of great stuff one can do with a nice video representation, e.g. some nice 3D finetunes :) Code: github.com/Stability-AI/g… Paper: stability.ai/research/stabl
GIF - The SD3 paper is finally out! Key findings * Reweighting noise sampling for improved rectified flow training * Improved transformer-based diffusion arch (MMDiT) * Scaling analysis for rectified flow transformers A lot of neat samples too, ofc! Enjoy! 🍿 bit.ly/3InY9Rd
- We have replaced our report on RDMs with the #NeurIPS2022 camera ready version. We show additional results for AR models, end the dominance of GANs on FFHQ and apply RDMs to zero-shot stylization. Thanks to @robrombach @kaan_oktay @jdotmueller B. Ommer. arxiv.org/abs/2204.11824
- I'll present our #CVPR2021 paper Understanding Object Dynamics for Interactive Image-to-Video Synthesis today (22.06.) 11am EDT/5 pm CET. J/w @timoMil @mdorkenw and B. Ommer arxiv: arxiv.org/abs/2106.11303 project page: bit.ly/3cxfA2L Join Q&A: eventscribe.net/2021/2021CVPR/…
GIF 
- Have you ever looked at a picture and wished you could bring it to life? If so, you should definitely check out our new paper ‘iPOKE: Poking a Still Image for Controlled Stochastic Video Synthesis’. arxiv: arxiv.org/abs/2107.02790 Project page: bit.ly/3dJN4Lf
GIF - Our work 'iPOKE: Poking a Still Image for Controlled Stochastic Video Synthesis' has been accepted to #ICCV2021 @ICCV_2021. Huge thanks to @timoMil , @mdorkenw and Björn Ommer. Start poking and check out our new colab GUI available in our code repoHave you ever looked at a picture and wished you could bring it to life? If so, you should definitely check out our new paper ‘iPOKE: Poking a Still Image for Controlled Stochastic Video Synthesis’. arxiv: arxiv.org/abs/2107.02790 Project page: bit.ly/3dJN4LfGIF
- Have you ever looked at a picture and wished you could bring it to life? If so, you should definitely check out our new paper ‘iPOKE: Poking a Still Image for Controlled Stochastic Video Synthesis’. arxiv: arxiv.org/abs/2107.02790 Project page: bit.ly/3dJN4LfGIF
- We are presenting "Semi-Parametric Neural Image Synthesis" today (11am-1pm) at #NeurIPS2022. The work shows the benefits of conditioning diffusion-based generative image models on visual instances retrieved from an explicit database.
- Check out our new work on human behavior transfer which is accepted at #CVPR2021. Joint work with @timoMil , @mdorkenw and Björn Ommer Arxiv: arxiv.org/abs/2103.04677 Project Page: compvis.github.io/behavior-drive… (1/2)
00:00 - Replying to @andi_blattOur model learns a representation of human dynamics independent of individual poses, enabling transfer of observed behavior to arbitrary start postures and also sampling of novel human dynamics. (2/2)
