or this one 🤔

Flux Kontext via Replicate Kontext just put every photo restoration company out of business. This took 6 seconds.

🚨BREAKING: China just launched the future of video creation. This AI tool literally creates motion design for videos in seconds with just one prompt. Here’s how it works (with real examples):👇

Sunlit Haven #3d artwork complete in #blender

Quick tip for anyone using multi image reference with Flux Kontext. If you have someone holding a product, you need to specify how big the product needs to be so that it looks accurate in their hand/s. By default the products appear massive from what I've noticed (unless you

Flux Kontext + Wan Vace in ComfyUI = a crazy combo! 🤯🔥 I used Flux Kontext through fal

Excellent news for Adobe Users

You certainly don't see that every day. Hailuo AI Director model

A little experiment I wanted to try with Wan Vase, Flux.1 Kontext, and Act 1. ❗️Disclaimer: Posting this is not me saying it would be better in color, that it should be made in color, or that this would be suitable for final production use. I just wanted to test what’s possible.

Street pose, neon shows

Low-Rank Head Avatar Personalization with Registers TLDR: Builds on the DINO-based zero-shot models (such as GAGA) by allowing for LoRA fine-tuning. This work adds a register module (per-vertex features + CNN) which improves fine-tuning capabilities for details such as wrinkles

One image, one virtual idol. Thanks, HeyGen, for the invite to create this "Vibin'" rap MV using Avatar IV. Unprecedented digital human product: - Flawless lip-sync for fast rap. - Natural hand gestures, rhythm-matched, no finger errors - One-minute single take. This is insane.

📢Code Release of Pixel3DMM 📢 Looking for a robust and accurate face tracker? Our state-of-the-art tracker handles challenging in-the-wild settings, such as extreme lighting conditions, fast movements, and occlusions. 👨‍💻github.com/SimonGiebenhai… 🌍simongiebenhain.github.io/pixel3dmm/

One of my favorite early tests with Luma modify video was just animating a rough mask in premiere and then photoshopping in a gnome to walk out of the shade and into the light.

HeyGen新的数字人Avatar IV有几个出人意料的亮点： 1. 最长60秒的说唱直出，无间断 2. 支持提示词精确控制手势 3. 高频歌词下，嘴部动作完全匹配 一张图就搞定了，对于想做简单mv的人来说，这很合适。

ByteDance Seed is back with SeedVR2, now on Hugging Face! This one-step video restoration model leverages diffusion adversarial post-training for impressive results, even on high-resolution videos.

Randomly Picked

I'm impressed with how well Veo 3 handled this prompt: > The video depicts a half length view of a beautiful young woman with subtle brown eyes, and dark eyebrows, she is standing by a wall. A movie is being projected directly onto her. The movie is of an old and beautiful

new chinese lip-sync method called OmniSync: