- Sink In
- Posts
- FLUX.1 Tools and New Video Models
FLUX.1 Tools and New Video Models
Long time no see - hope you still remember us, the SinkIn Newsletter, a 5 minutes read made at sinkin.ai to cover the most interesting stuff in the Image AI world.
BlackForestLabs has rolled out FLUX.1 Tools—an advanced toolkit built on their FLUX.1 text-to-image model. These new features let you seamlessly edit, reshape, and restyle images without losing quality:
FLUX.1 Fill: Easily inpaint and outpaint to edit or extend images.
FLUX.1 Depth & Canny: Maintain structural details through depth and edge maps, letting you tweak textures while preserving a photo’s core layout.
FLUX.1 Redux: Generate image variations and combine images with prompts for fresh, high-quality visuals.
Flux Inpainting
Tencent has released Hunyuan, a new open-source AI video model capable of generating high-resolution, 5-second videos from text prompts. Early tests show impressive visuals and motion, comparable to top commercial models. This 13-billion parameter model is a significant addition to the open-source space, allowing community-driven improvements. While demanding in its current resource requirements, its open-source nature and strong output quality make it one to watch.
Video Generated by Hunyuan
Made by Lightricks, LTX Video is a text-to-video model that generates high-quality videos in real-time. It can generate 5 seconds of 24fps video at 768x512 in just 4 seconds on an Nvidia H100. Our test shows it could take quite some efforts to generate content that’s satisfying, but it is indeed faster and cheaper than other top line video models such as Kling or Sora. You can try it out here.
World labs, led by the Godmother of AI Fei-Fei Li, unveiled their first step towards spatial intelligence: an AI system that generates 3D worlds from a single image. This lets you step into any image and explore it in 3D. It could be a game changer in game development, video production and virtual reality creation.
Image to Interactive 3D World
Similar to Stable Diffusion, people have been working on Flux finetunes too. UltraReal is a Flux finetune that tries to push realism to the next level, finding that sweet spot between amateur aesthetics and professional, high-quality visuals. The author has just released 2.0 which was trained on expanded datasets and 205,560 steps. It offers better hands / feet / poses, sharper textures / quality, and improved text rendering.
Meme of the Day
That’s it for today, hope they are as warm as a cup of hot chocolate!
What'd you think of today's edition? |