Sand.ai

🌟 Highlights: - Blackwell Support: Early Blackwell support via FFA_FA4 backend, leveraging HSTU function & R2P optimizations for next-gen hardware. - Native Group Collective: DeepEP-inspired fused kernels (GroupCast/GroupReduce) to break RDMA bottlenecks and achieve zero-redundancy communication. - Mask-Agnostic SOTA: Constant high-throughput performance even for irregular, complex masks. No more OOM or performance drops in "hard mode." - System-Level Synergy: Dispatch Solver + Adaptive Overlap ensures near-linear scalability on H100 & B200 clusters.

🪄 Introducing MagiAttention v1.1.0: Defining the Performance Ceiling for Hopper & Blackwell Architectures. MagiAttention is a distributed attention engine sharing the bottlenecks of ultra-long context and heterogeneous masks. - Repo: github.com/SandAI-org/Mag… - Blog: sandai-org.github.io/MagiAttention/…

Thanks for reading! 🙌 Come visit us at Sand.ai to see what else we're building. We are incredibly proud to contribute to the open-source community and will keep pushing the boundaries of AI! ⏳ 6/6

📊 Performance & Benchmarks: Beating Open-Source SOTA Great architecture is nothing without results. We benchmarked daVinci-MagiHuman against leading models like LTX-2.3, Ovi 1.1, and MoVA. The results speak for themselves: 👥 Human Blind Evals: Tested on a 100-sample dataset across multiple dimensions, achieving a massive 70.5% overall win rate! 📈 Objective Metrics: 🎬 Video (VideoScore2): Outperforms LTX-2.3 in Visual Quality & Text Alignment; superior to Ovi 1.1 across the board. 🎙️ Audio (TalkVid-Bench): Crushes the competition with vastly lower Word Error Rates (WER) for crystal-clear, accurate speech. ⏳ 5/6

🪄 Introducing daVinci-MagiHuman: The Performance-Level Audio-Video Generative Foundation Model Proudly open-sourced and jointly developed by SII GAIR Lab & Sand.ai, it sets a new standard for multimodal AI. ⏳ 1/6

The full version of Gaga-1 is here. 🎙️ Voice Reference 📱 Both 16:9 & 9:16 Video ✨ 1080p Export Free for all. Create now.

GAGA-1 is out now. Free for all. No invite needed.

GAGA-1 is coming. The AI actor with voice and visuals as one. 🎭 Hollywood-level emotion in every frame 🎬 One-click audio and video sync 🌍 Perfect performance in multiple languages 👇 Here are 9 wild examples:

🪄 Magi-1.1: The Most Powerful Video Extension - Now available at sand.ai/magi ♾️ Make your video as long as you want ⏭️ Every move flows perfectly into the next 🔗 Everything in your shot stays exactly the same ⏳ 1/3

We understand your concern. We cannot confirm if unofficial sites use our API. Importantly, these sites existed before we offered any public API access. We are actively working to take them down through complaints and legal action. For the official, secure service, always use Sand.ai/magi.

@SandAI_HQ @magiai_hq Ok so my money didn’t go to you but rather someone that had an API to your site?

I hate chargebacks, but... This is probably the WORST video AI tool I've used yet. @magiai_hq @SandAI_HQ So, what makes them so bad? ...Well, first they give you 16 free credits to generate a sample video. ...Turns out you can't generate a sample video. I decided to just pay the $30 on a monthly plan. Here's the output... If you take a look at the link in the comments, hopefully it saves you $30. I emailed them to get a refund. But curious if anyone's taking bets as to whether or not I would have to issue a chargeback, which coming from e-com, I hate, hate, hate doing.

🎬 Gaga AI: A tool that animates any photo into a lifelike, talking Character. - Try it for free at gaga.art Key Features: 🗣️ Pro-Level Realism: Get perfect lip-sync, natural facial expressions, and even hand gestures for your Characters. 🎭 Emotion Control: Let your script breathe life into Characters, turning a simple “Happy” into a genuine smile. 🪄 Simple & Intuitive: A clean UI designed for a smooth, easy, and fun creation process. Whether you're a content creator, educator, or marketer…, Gaga AI helps you produce amazing content fast.

🪄 @GagaAI_official is now on X.

@giffmana You can take a look at Magi-Attention, which supports arbitrary attention masks and delivers performance comparable to Flash Attention 3, along with built-in distributed capabilities. x.com/SandAI_HQ/stat…

So I have these big fat `triton_tem_fused_zeros` in my backward pass profile. They take up about a quarter of my steptime, more than matmul(!) After some time, I found out via elimination that they are related to flex_attention: if I replace x = flex_attention(q, k, v, ...) by x = q+k+v they disappear. Any flex_attention expert here knows about this? @cHHillee I know you don't work on it anymore, who would you say is the person to ping on flexattn Qs nowadays?

🎨 MAGI-1 now supports ComfyUI! Try it out now! 📖 User Guide: github.com/SandAI-org/MAG… ⏳ 2/2

📢 Introducing the Sand.ai Open Platform: access Magi's top models via Open API. 🎬 Get started: platform.sand.ai/docs/get-start… 🎮 API Platform Console: platform.sand.ai 💰 Price: 0.5$/5s only. ⏳ 1/2

🪄 Announcing 4.5B Magi-1 Distilled and Quantized Models - Code: github.com/sandai-org/MAG… - Checkpoint of Distilled Model: huggingface.co/sand-ai/MAGI-1… - Checkpoint of Distilled+Quantized Model: huggingface.co/sand-ai/MAGI-1… 🌟 Highlights: - A 4-second 720P video can be generated in just 8 minutes on a single GTX-4090 GPU with our quantized model - Maintains the same performance as the 4.5B full model with only 1/12th of the inference cost