Pinned Tweet
Every new AI video model promises cinematic quality, but consistency is what really matters.
This was created with Seedance 2.5 on Morphic, and the motion, camera work, and visual quality stayed impressive from start to finish.
If this is where AI video is heading, creators are about to unlock a whole new level of storytelling.
@morphic
Morphic.com
Alexander Jackson retweeted
Baidu just launched ERNIE 5.5 Flash Preview, and what it's capable of is pretty incredible.
Take game development, for example. The model built nearly 100 playable mobile games from game design documents.
And no, I'm not talking about simple code demos.
It was given complete game specifications and tasked with building games across different genres, from tower defence and racing to card-based RPGs and interactive stories.
I went through the demo, and a few things caught my eye:
1. It actually builds playable games.
Not just the basic mechanics. It handles the artwork, sound effects, gameplay logic and other pieces needed to make a complete mini-game.
2. It isn't locked into one framework.
The 24 games featured in the demo were built using Cocos, Unity, LayaAir, PixiJS and native Canvas 2D.
That's quite a range.
3. It can test its own games.
This is the part I found particularly interesting.
For most of the games, the model uses simulated players, state-space search and level replays to check whether levels can be completed and whether the gameplay is balanced.
Basically, it doesn't just build something and leave you to figure out what's broken.
4. It pays attention to the little details.
Different games get their own character names, boss dialogues, branching storylines and level names tailored to their genre.
And there's more to ERNIE 5.5 Flash Preview than game development:
- 280B total parameters, with 14B active per token.
- 1M-token context window.
- Generation speeds of 200+ tokens per second.
- Support for complex coding and multi-step AI agent workflows.
- Knowledge work across 40+ file types.
- More than 1,000 creative skills covering writing, visuals, video and production.
- Code Arena: WebDev score of 1,668, trending toward #2 among open models and ahead of DeepSeek-V4.1-Flash (Max) at 1,664 and Kimi K3 (Max) at 1,658.
The API is already available through Baidu's Qianfan platform, and the model weights are scheduled to be open-sourced before November 25.
My take:
We've seen AI generate code for a while. But watching it go from a detailed game specification to a playable game, across different frameworks, and then testing whether the game actually works is a different story.
I'm curious to see how far this can go.
#AI #GameDevelopment
Alexander Jackson retweeted
Every new chat with an AI assistant is a small restart.
Your context has to be rebuilt from scratch, every single time.
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience.
On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
Alexander Jackson retweeted
You don't always need a bigger production budget to make your marketing content look more put together.
Start with Dreamina Marketing Studio to generate product visuals and creative directions.
Then bring your materials into CapCut PC and shape them into campaign videos.
Dreamina: tinyurl.com/376c8r3x
CapCut PC: tinyurl.com/2v23rn22
#CapCutPC #Dreamina
🤝 Paid partnership · 🤖 Made with AI
Alexander Jackson retweeted
Side-project builders, try giving your agent a short memory file and timing the same job before and after. Easy way to see if memory is doing anything.
🤝 Paid partnership
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience.
On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
Remembering a task and getting better at tasks sound like the same achievement. They're quite different, and AI makes that very clear.
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience.
On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
Alexander Jackson retweeted
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience.
On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
Alexander Jackson retweeted
One product shoot only goes so far when you need multiple Black Friday ads on a tight budget.
Dreamina Marketing Studio helps you generate the initial product visuals and campaign concepts. Then CapCut PC makes those materials work harder: AI Edit changes selected visual details, while AI Extend generates additional footage for longer shots.
Need to polish the final video? CapCut Agent EditPilot lets you describe editing changes in simple language, making it easier to refine your content without handling every adjustment manually.
Reuse and adapt your creative assets instead of planning a new production every time.
Dreamina: tinyurl.com/376c8r3x
CapCut PC: tinyurl.com/2v23rn22
#CapCutPC #CapCutDidThat #Dreamina #Smallbusinessmarketing #chatwithcapcutagent
Alexander Jackson retweeted
🚨 You can now build your own AI second brain that stays 100% private. In 2026, you can now build an AI second brain in 5 steps.
Karpathy popularised the idea of having a second brain powered by AI.
But imagine having an actual app that you own, where your knowledge stays PRIVATE.
Here's how it works:
1️⃣ Store all your notes as plain Markdown files on your own device.
2️⃣ Ask AI any question about your notes and get answers with citations pointing to the EXACT note.
3️⃣ Build a LIVE knowledge graph that automatically connects your ideas.
4️⃣ Connect it through MCP so Claude, Cursor & ChatGPT can access your entire second brain.
5️⃣ Use it across Windows & Mac without giving up ownership of your notes.
It's called Socratus.
Think of it as having your entire knowledge base accessible to AI whenever you need it.
Try it here. 👇 It's free to get started, with a Pro plan for power users.
socratus.app/
🤝 Paid partnership
tested one for a pretend guest house and I tried to confuse it and it just rolled with it lol
Introducing @FishAudio Live 3: a full-duplex voice model that listens, says "mm-hmm," and knows when to shut up.
Enterprise? Book a demo for your free custom agent: s.fish.audio/vxhgs6
Reply with your use case - we'll build it free for the first 1,000 people.
Alexander Jackson retweeted
Your Android phone has features Apple users can only wish for.
And here's the crazy part: most Android users don't even know they exist.
Here are 8 hidden Android features you need to try RIGHT NOW:
Alexander Jackson retweeted
I tried using AI to create an industrial robotic hand blueprint using MiniMax's H3-Image AI Model 🔥
I really love how it brings the mechanical details together.
Prompt Below 👇
I used H3-Image to add:
👉 a worn-out metal texture
👉 Articulated joints
👉 Black cables
And finally it exploded the assembly views to make it look like an actual engineering prototype 🤯
Honestly, This feels like something straight out of an industrial design studio. @MiniMax_AI
🤖 Made with AI
Alexander Jackson retweeted
Greek improved 77% on Scribe v3 Live vs v2 Realtime in their evals, Danish 54%. We caption live events in both
🤝 Paid partnership
Alexander Jackson retweeted
Can we see a few more options?' A sentence with its own villain arc.
Buy or upgrade to Ultra on CapCut Web during Nov 12-18 to unlock 7 days of unlimited image generation on select models.
Try a few creative directions, shortlist what fits the brand, and build from there. Let someone else explain what 'more premium' means.
capcut.com/my-edit?utm_mediu…
#CapCutWeb #CapCutWebDesignStudio #AIImage
🤝 Paid partnership
if your parent goes to the doctor without you
this is worth looking at.
an AI jewelry necklace that keeps what the doctor said so you're not piecing it together later.
I'm sending this to my dad.
AI can now tap your shoulder
This is Luna, and it speaks up before you have to ask
Back Luna on Kickstarter now: moonshot.computer/luna
Alexander Jackson retweeted
Your LiveAvatar just leveled up.
Any frame: close-up, half-body, full-body
Every feeling: 6 emotions
Now in 1080p
All from a single image.
Available now 👇
Not every small business has the budget for a full commercial production. That's why I like experimenting with ideas like this: a secret agent spying on a product target, sneaking into the scene, and finally enjoying the product.
I used a product photo and instructions in CapCut PC to create the video concept, then worked on the edit. Multi-Track Editing was especially helpful for keeping the background music, action sounds and product footage organised. More control, less time searching for the right layer.
See how it works: tinyurl.com/2v23rn22
#capcutpc #capcutmarketing #capcutblackfriday #CapCutDidThat #capcuteditpilot
Alexander Jackson retweeted
Introducing the HeyGen Platform and Playground.
The same models behind HeyGen, now raw and in your hands. Try HeyGen Video, HeyGen Voice and our avatar models, then take them straight to production through the API.
platform.heygen.com
Alexander Jackson retweeted
My UGC presenter says basically so often that it should have its own product page.
Clean up your talking clips in CapCut PC.
tinyurl.com/2v23rn22
#capcutpc #capcutblackfriday
🤝 Paid partnership · 🤖 Made with AI
Alexander Jackson retweeted
An art director’s next question: what can we do with this image now?
Seedream 5.5 offers:
› independent raster assets from scene decomposition
› foreground extraction with background reconstruction
› native transparent output
› continued edits on the same draft
The handoff from generated image to design material matters.
The image price, at a third of GPT-image-2.6’s, gives the production team something else to consider.
What happens after 100+ rounds of “just one more change”? How far can you push an image before it starts to drift?
Meet Seedream 5.5: precise local edits, consistency across repeated revisions, more natural visuals and native transparent assets.
Explore what’s new below. 👇
byteplus.com/en/contact-us/s…



















