Okay, Claude is the real deal. 🤯
I handed Claude Opus 5.5 our Sumi Studio MCP and it directed this whole 45-second Halloween motion piece by itself 🎃 12 scenes, 197 sound cues, every hit on the beat.
This one workflow could crank out brand videos and explainers all day, and spin up new accounts on other platforms fast. @claudeai you cooked. Sound on 👇
My favorite thing about working on a canvas: when shot 4 comes out wrong, I regenerate shot 4. The character, the other shots and the cut stay put.
Our team wrote up how the @sumistudio_ai canvas works, from the first node to the final edit.
New to the Sumi Studio canvas? Here's how it works: adding nodes, connecting shots, fixing one scene without touching the rest, and editing the final cut, all in one place. https://t.co/pAdEYru5Wh
Street Fighter's Cammy, off the page in 44 seconds.
She was dots. Then lines. Then a comic panel. Then flesh. Then she walked into a soulslike boss fight.
Based on @MANISH1027512's Cammy cos 💚 Made with an agent on @sumistudio_ai
🧟♀️ MJ's Thriller, redone with 9 AI-model girls: DeepSeek, Claude, GPT, Grok, Gemini, Kimi, GLM, MiniMax and Qwen. For Halloween 🎃
Made with Codex + @sumistudio_ai. Characters by @ZipZipPipe (CC BY-NC-SA 4.0). This video is shared under the same license.
132s, all 54 original shots, original timing. How it was made below 👇
Okay, Claude is the real deal. 🤯
I handed Claude Opus 5.5 our Sumi Studio MCP and it directed this whole 45-second Halloween motion piece by itself 🎃 12 scenes, 197 sound cues, every hit on the beat.
This one workflow could crank out brand videos and explainers all day, and spin up new accounts on other platforms fast. @claudeai you cooked. Sound on 👇
@KurmaRitish@fuguisanqi Chaining ~15s beats is the safer bet. One long prompt asks the model to plan the whole edit, and that's usually where it slips.
@kheisler2@sumistudio_ai Fair, the misses are where the credits really go. I draft short and low-res until the motion lands, and only upscale the take that makes the cut.
Genie 4 just dropped (testers only). Honest take: I'm bearish on world models. Huge compute, meh UX, no use at scale yet.
But Gemini directing the world between inputs is basically a DM. So I mocked up a D&D table. Made in @sumistudio_ai, not Genie footage. Sound on 👇
This is Genie 4, our most realistic and interactive world model yet. 🌐
Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak.
@azhui2026 Keeping the prompt next to every clip is the part that saves you. Most of my wasted retries come from forgetting which version actually worked.
@matotoLab@sumistudio_ai@ZipZipPipe The model decided Thriller needed an axe and I wasn't going to argue 😂 Halloween rules: whatever shows up in the take stays in the cut.