☀️☀️ Sora2 Face Inconsistency? Here’s My Flora-Powered Workaround☀️☀️ Sora2 does not allow realistic human faces due to its policy restrictions, which makes it difficult to maintain visual consistency for a character across different scenes. Here’s how I solved that, using Flora as the hub: I first design the spot, then I generate a consistent model/character. To maintain coherence across angles, I lean on Kling or Veo3, which help preserve the same “look” from different perspectives. Alternatively, you could extract the initial model’s frame, then reimagine her in different positions or settings by using Seedream or Nano Banana, and finally animate her. It’s a pipeline that gives continuity without losing realism — especially useful for spots, short films, or narrative pieces with recurring characters. A full FLORA tutorial is coming soon 🎥 If you want to try FLORA now, here’s the link 👉 🩷 🩷 My Link get 25% off for 12 months https://lnkd.in/dgKcsupm And if you like watching my AI experiments, hit Follow — more to come. #alessandrabalzani #tutorial #sora2pro #florofaunaai #kling #veo3

What the hell is an LLMS.TXT file? It sounds confusing but it just acts as a guidepost—helping LLMs quickly locate, understand, and reference your most important content. Today we talk about why it matters for businesses and publishers, and how it shapes the way your content interacts with AI systems. https://lnkd.in/eSvQFBGa

Sora 2 is absolutely insane. Credits for the video - Albert Bozesan This new version takes short-form, audio-matched generative video to another level physics, better prompt adherence, and more control than the first model. For quick reels, animated shorts, and content made for social feeds, it’s starting to feel production-ready. it's aware of context, it understands backgrounds, scene logic, and details. It feels like a different beast entirely. Where It’s Still Weak - Long-range storytelling, multi-shot flow Text rendering, hands, and small objects Guardrails + safety controls Where It’s Already Strong - Sharp, believable short clips with audio in sync Flexible inputs (prompt or image) Horizontal + vertical outputs with dialogue and SFX More realism. More accuracy. More control. This one feels like the closest we’ve been to usable generative video yet #openai #sora2 #aivideo #genai #sora

Continuing my exploration of AI video and motion design. The previous test was created in After Effects using particle simulations... this one blends Google Veo with compositing to explore how digital dust can feel natural. The concept product is an exfoliating powder, where each particle represents renewal ... light, air, and mineral texture forming calm I still feel this could look even more refined if crafted fully in After Effects, but it’s fascinating to see how close AI can now come to the softness of physical materials. I’m still wondering which feels more tactile to you, the AE particle version or this AI-driven reveal? #AIvideo #CreativeDirection #AfterEffects #Veo #DesignAlchemy #BeautyDesign #MotionDesign #BrandStorytelling

So with AI Video production starting to get more and more realistic, the team at Creative Kiwi have been slowly upping their game when it comes to video. Its starting to be a crazy time to live in right now! Check out some of the work that they have done

Veo 3.1 is impressive! Took the Ingredients feature in Fast for a spin, using the isolated elements (Alien on an alpha channel, Newscaster on a white BG and the news set cleaned of any characters with Photoshop generative fill) of what would have gone into my Runway References workflow back in late May when I originally made 'The Visitor'. Very impressed with the consistency this maintained! Audio generation is much cleaner so far, free of the heavy artifacting that came with most generations. Also the humour of the performance of the anchor peering around... nice touch! Prompt: Make the Alien in image 1 sit at the newsdesk on image 2. The News Anchor in image 3 is peering around a background set corner at the alien in disbelief. The Alien is talking about ways to bring about world peace instantly. Note that Ingredients is only available for Veo 3.1 Fast directly through Google's Flow platform.... which is quite worth the Ultimate price tag if you're doing high volumes of generations (you can generate 20 clips at once!), as Fast generations are still unlimited! More to come as we learn more about this new beast of a model. #alien #cliffhanger #veo3 #google #deepmind #test #consistency #film #ai

🎞️ The “Infinite Scroll” for Video Is Here — and it FLIES! 🚀✈️ Just taught my video generation agent to do frame chaining → now it can create essentially infinite continuous shots. 🎬 This works with any img2video model. 👉 Example: this flight scene built in Seedance, starting from a single image generated in Imagen 4. ✨ Why it’s a game-changer: Seamless agent workflow (no manual stitching) Frame chaining (end frame → start of next = endless flow) Paired with AI audio (like Suno 5) → cinematic impact 💡 The takeaway: The future of video isn’t short clips, it’s fluid, never-ending narratives.

Our MD, Ben Stallard, introduces you to Slapshot, the AI-powered VFX toolkit designed to make advanced effects fast and intuitive. With tools such as: Rotoscoping Generative Retiming Video Depth Maps Camera Tracking Slapshot puts powerful VFX workflows at your fingertips, so you can focus on creating.

Solution Spotlight: Slapshot AI Camera Tracking Revolutionizing VFX workflows with AI-powered precision Slapshot, developed by Hotspring, is changing the game for VFX artists and compositors. Traditional camera tracking can take hours or even days. Slapshot solves it in under 10 minutes—no metadata required. What’s New? - AI-driven camera solves with accurate point clouds - Supports .abc, .glb, .exr STMaps, .jpg, .mov - Runs securely on AWS, up to 8K resolution - Designed for 2D artists—no 3D expertise needed Industry Buzz: Shortlisted for Best AI Innovation and Standout Tech of the Year at the 2025 Broadcast Tech Awards. “Slapshot isn’t meant to replace expert tracking on heavy CG shots—but for medium-difficulty work, it’s a game changer.” — Jon Mason, CEO, Hotspring #VFX #AIinMedia #CameraTracking #PostProduction #SlapshotAI #BroadcastTech #MediaInnovation #CloudVFX #CompositingTools

I’ve been testing a video inpainting workflow using crop and stitch. It’s working well, just not quite perfect for V2V lip sync. Still a work in progress. This is also my own AI music and AI characters and video. Since ControlNet can’t be added during long-gen Infinite Talk, the new frames don’t match the original 100%. I’ve been compensating by lowering the denoise, but of course that affects lip accuracy. There is also this ignoring jitter that's from the mask, which needs some type of temporal smoothing, if anyone has any ideas. The big upside is that it only generates the cropped region, completely ignoring the rest of the high-res video. That makes it fast and resource-friendly and I think it’ll end up being an even better solution for aging faces, which has always been one of those costly, time-consuming VFX challenges. Stay tuned for that one!