sada

Sada | AI Art & Visual Creator ✨
Exploring boundaries of generative imagery &video & prompts.
#AI #AIArt

[sada @sada_ai]

This AI-generated "Fat Frog" animation is absolute gold! 🐸💥
Created with Seedance 2.5, the prompt reads like a cinematic storyboard. Love the "leaf sunglasses" and the charred, shocked expression after the explosion

[sada @sada_ai]

Don't scroll past this! This is the most hardcore Halloween visual I've seen this year...

[sada @sada_ai]

@emollick Amazing

[sada @sada_ai]

DeepMind's Genie rewrites game dev: ride a dragon, WASD-control camera/flight, type 'castle launches fireballs' and physics responds in ms. Before: years of C++ and manual art. Now: zero code, prompt-to-world in seconds, real-time edits. Long sessions still drift, but everyone gets an AAA world.

[sada @sada_ai]

This demo by Justine Moore is insane. That's not a game recording—it's a fully interactive, real-time video generated by Genie 4.

Look at the UI: running at 24 FPS, you can actually use the prompts at the top to change the world on the fly—like switching to "Wet Pavement Rain" or "Sunset Skyline Rays" in real-time. The Waymo car drives through the neighborhood with instant physics and lighting calculations.

[sada @sada_ai]

Don't just view Google's Genie 4 as a flashy "AI game" demo. From a tech perspective, it's a massive leap toward building true physical world simulators.

Unlike video models obsessed with pixel quality, Genie 4's killer feature is real-time interaction and physical feedback. The 2-minute consistency means it actually "remembers" past events, with physics (gravity, collisions) feeling intuitive. Plus, native real-time audio lets characters talk back to you.

[sada @sada_ai]

Just saw the Google DeepMind Genie 4 demo, and honestly, every other video model just became obsolete in my eyes. 🚀

This isn't just "generating video"—it's generating an entire world you can step into, interact with in real-time, and actually hear. Sundar Pichai called it their "most realistic and interactive world model yet," and after watching the demo, I totally believe it.

[sada @sada_ai]

Most people using Seedance 2.5 are still writing long prompts and tweaking parameters. But there's an underrated capability: use a reference video + a reference image to directly replace the person or scene in the original video.

No generating from scratch, no describing camera movements, no complex prompts. You only need two things: an existing base video, and a photo of the person you want to put in it.

The workflow is simple. Upload the base video as the "video reference" and the person's photo as the "image reference." The prompt only needs three sentences:

"The person should only reference Image 1. Replace the person in Video 1 with the person from Image 1, keep everything else unchanged. Keep the colors consistent with the original video."

The first sentence prevents the model from dragging in the image's background too. The second is the key—it tells the model to swap only the person, while the camera, motion, lip sync, and pacing all follow the original video. The third locks the color tone, avoiding a washed-out unusable clip.

From testing, pick a base video with a single person, a clear face, and not-too-fast motion—trim it to around 10 seconds for the lowest cost of trial and error. For the person's photo, a three-panel image works best: front headshot, full-body front, full-body back, with only one face in the entire image. More faces will distort the swapped person.

Seedance 2.5 itself supports up to 30 images and 10 video clips as reference material per generation. Reference images aren't charged extra—at 720p it's 6.10 credits per second.

Before posting, remember to check the AI-generated content label. Use base videos you shot yourself or have licensed—don't scrape someone else's viral clip. Your account matters.

[sada @sada_ai]

I Wanna Love You Video recast

[sada @sada_ai]

Seedance 2.5 4k is coming

[sada @sada_ai]

The videos made with this Seedance 2.5 are really high quality.

[sada @sada_ai]

Amazing video

[sada @sada_ai]

recast video with ai

[sada @sada_ai]

seedance 2.5 video

[sada @sada_ai]

thisVvideo is create with Seedance 2.5

[sada @sada_ai]

This video made with Seedance 2.5 is pretty good.

[sada @sada_ai]

@AIwithkhan @CuddlerAI That's fucking scary.nice work.

[sada @sada_ai]

Kimi K3.1 is the Opus 5.5 / Astra killer,do you believe？

[sada @sada_ai]

Kimi 3.1 Pro is Great and cheap.

[sada @sada_ai]

totally done in by this cat, haha. The prompt is below.
Real raw phone video filmed by a neighbor from a porch across the road, one continuous 30-second take, no cuts. Fern top left and porch railing bottom left stay in frame. Handheld wobble, clumsy pinch-zooms that go soft and blocky. Golden-hour sun, compressed phone quality.
[IMAGE REFERENCE]
the old cream-and-brown pickup truck with a cargo bed in side profile on a steep grassy slope, nose uphill on the left, rear downhill on the right toward the lake, "LAKE LIFE" sign, a piece of firewood wedged under the downhill side of the single tire. Three heavy-set Alabama men in lawn chairs in the truck bed with a blue cooler.
[TIMELINE]
0-4s: The men clink cans and drink. Neighbor drawls: "Y'all are gonna die up there." Man in the ALABAMA tank top: "Hold my beer!"
4-12s: Clumsy zoom to the tire. A black house cat walks out from under the truck, bats the firewood, drags it out like a toy. The truck creaks; the cat flattens its ears and bolts uphill out of frame, never seen again.
12-20s: Zoom out. The truck starts rolling backward down the slope, speeding up and bouncing. The men grab their chairs yelling "WHOA WHOA WHOA!" Neighbor: "Oh no, OH NO!"
20-23s: Just before the water the men look at each other, shrug, sit back and raise their cans as the truck plunges rear-first into the lake with a big splash.
23-30s: Zoom in. The truck sinks slowly. The men stay seated, calm, open fresh cans and toast as the water rises to their waists. The bearded man pulls a phone from his overalls and the three pose for a selfie, cans up, the sinking truck behind them. Ends on three heads above the water, cans and phone held high.
[AUDIO]
Phone mic only: crickets, cans, laughter, the lines above, cat paw on wood, metal creak, rolling and scraping, splash, bubbling water, cans cracking, a camera shutter click, neighbor laughing hard. No music.
[QUALITY]
Photoreal raw phone footage, identity locked to image1, realistic pickup-truck-with-cargo-bed weight, rolling and water physics, realistic cat. Nobody gets hurt. No cinematic look, no slow motion, no text, no cuts.