Speculative ad brief for DefendUAS. Shifting the needle from basic product animation to high-stakes military storytelling.
Visually contrasting a gritty, hyper-realistic urban terrain with sharp, tactical holographic wireframes to show exactly how a distributed mesh network reclaims the decision window. Using a scene-by-scene strategy in Melius, starting with a master script that fed into storyboards and hero keyframes for total visual continuity across the cinematic clips. This workflow allowed me to lock the moody tactical aesthetic before generating the final video scenes, deep-voice narration, and orchestral score. Everything is laid out from left to right, showing the journey from concept to final assets. The project screenshots/canvas video are in the comments.
1
3
711
I created THE TABLE because the most uncomfortable human moments deserve to be witnessed without the mercy of a camera cut.
We have all seen dramatic cafe encounters on film, but they are usually softened by edits and reaction shots. I wanted to create an experience that traps the audience in a continuous reality, making the viewer feel the tension of a silent conflict unfolding in real time.
What if we watched a bizarre fracture between two people in a cozy space, and the camera simply refused to look away? That question became THE TABLE.
The Production Workflow
I had a vision but no traditional crew. CapCut Video Studio and the Seedance 2.5 model made a complex one-take sequence entirely possible.
Prompt-Driven Choreography: I mapped out a 30-second physical space in a single text prompt. Every action, from the walk to the aggressive muffin snatch, was engineered to unfold sequentially.
Unbroken Camera Control: To ensure the "one-take" rule was followed, I instructed the AI to use physical camera movements. Seedance 2.5 maintained the spatial consistency flawlessly.
The Assembly: The magic of this workflow was the lack of assembly. The final output is a raw, continuous generation with no manual splicing.
Platform Feedback
Seedance 2.5 brilliantly held the entire world together. When the camera whip-pans back from the street to the interior at the end, Table 1 and Table 2 are exactly where they should be. It felt like directing a steady-cam operator on a live set.
One honest piece of feedback: generating timed character interactions in a continuous 30-second shot requires immense trial and error. Optimized rendering costs for long-form visual storytellers would make a meaningful difference, also if specific editing can take place at point notes that will change massive things.
Ultimately, Seedance 2.5 helped me protect the awkward silence at the end of the film.
X: https://x.com/AshishSandhu/status/2090161080857223462?s=20
4
7
414
Ever tried to close one notification and 10 more pop up instantly?? 😵💫🏠 "Apartment for sale", "Special offer!", "Last chance deal!"... the screen never stops! 😂📲
🎬 Visual experiment with AI for TAPU Real Estate Agency
🔁 Chaos + notifications + confusion = the real mood of 2026 💀
👇 Which notification chases YOU the most on your phone? Drop it below! 😅
1
136
My Oura ring knows how I slept, but it can't see my bedroom. So I built a small box with an ESP32 and four sensors (CO2, temperature and humidity, noise, and a magnet switch on the window) that writes the room down every two minutes. A Google Apps Script lines every 5-minute point from the ring (heart rate, HRV, sleep stage) up with the room readings from the same minutes, and every morning I get one report in Telegram.
Code does all the maths. NVIDIA Nemotron, running on Nebius Token Factory, only turns the facts into a message a person wants to read, and code checks every field before it is sent.
What seven nights showed: on my hottest night (23.1 °C) my HRV was 32 ms against 42 on my other nights, and my pulse was 69 against 60. Seven nights is a hint, not proof.
0
47
A series of AI-generated lifestyle images and short-form videos designed to replicate authentic, customer-style content for a craft-focused e-commerce brand.
Starting from scratch, I created an original patchwork tiger plush and built a small “brand world” around it — focusing on believable, everyday moments rather than polished campaign visuals. The aim was to produce content that feels natural, imperfect, and genuinely user-generated.
The work spans unboxing, material selection, and home interactions, extended into short-form video clips that capture simple, relatable moments of use and discovery. Particular attention was given to lighting, composition, and subtle imperfections to avoid a staged or overly commercial aesthetic.
Created using Nano Banana 2 (image generation), Kling (video), and iterative prompt development to refine realism, continuity, and behaviour across both static and motion assets... Exploring the intersection of generative AI and performance-led creative.
0
39
A movie-style close-up of a vampire holding a melting ice cream in his left hand. He slowly and delicately licks the ice cream to savor its flavor never biting it. His tongue gently touches the soft scoop, as the ice cream is light and airy, meant to be enjoyed only by licking.
A bat rests on his shoulder, calmly watching and leaning closer to sniff the ice cream without interrupting him.
The camera is slightly shaky, creating a natural handheld feel. In the background, the amusement park is alive with movement (no visible people), and the carousel on the left rotates left from the camera’s perspective.
1
69
