Heba AI

@vadooai wan

🚨 WAN 2.7 Early Access Big upgrades: ⚡ Visuals • 🎬 Motion • 🔊 Audio • 🎨 Style • 🧠 Consistency First/last frame video, 9-grid image→video, voice + subject reference, editing & recreation. No waiting. No delays. Be first to use it. Follow, Repost & comment “WAN” for early access 🚀

Back in the day, photoshop tutorials was bought as cd-roms. This one by Anders Qwicker.

@OfficialLoganK U have issues to work first

Our AI Studio vibe coding roadmap for the new few weeks: - Design mode - Figma integration - Google Workspace integration - Better GitHub support - Planning mode - Immersive UI - Agents - Multiple chats per app - Simplified deploys - G1 support And more, should be fun : )

@francolli @BrentLynch $8 imagine was quite good value giving like 1000 videos a month with just $8. New model not so much.

@SubarcticRec @BrentLynch I get that but free with premium + on a site that is opaque and works on the whims of Nikita for $50 per month, or standalone without premium + $30 per month for a video tool that can't lip sync close to other models Mmmm, choices choices I'll take my chances with other models

😲GASP! The paywalls are going up. You'll now need a SUPERGROK or PREMIUM+ ACCOUNT (includes SUPEGROK) for GROK IMAGINE App Usage. At the moment you can still create low resolution images inside of Grok on X with a Premium account. While I'm a fan of Grok Imagine until they roll out all the features available via API (Key among them 15 seconds) I most likely will use Grok Imagine via API.

The reason why Google AI documentation is useless is they let gemini vibe code the documents. Gemini is mostly bad for everything.

@francolli @BrentLynch Its less censored than other closed models.

@BrentLynch Nah, Im done with grok imagine Its like sitting in a room filled with lego technics sets and all you are allowed to use is a bucket of duplo bricks For $30 a month Its shit

Graphic designers, look away. - Adobe just dropped a new AI model. - It generates fully editable layered files from simple text. - No more flat AI images. - Real RGBA layers with perfect transparency. - It even rips flat JPEGs back into separated layers. - The architecture is honestly beautiful. And... the open-source GitHub link? - no, it's Adobe. Enjoy reading the PDF. arxiv.org/pdf/2603.17965

These happen if not being careful. Agent went and overwrote all my work by pulling git checkout from some public repo.

@OfficialLoganK Maybe try to fix it so that Gemini 3.1 pro would not still be so outdated when using in AI Studio.

Spent all day vibe coding, fixing bugs in AI Studio, and polishing the experience :) So much fun!

@claudeai Also in Antigravity?

1 million context window: Now generally available for Claude Opus 4.6 and Claude Sonnet 4.6.

@ziwenxu_ I would not use OC for automations as there is better tools for that. OC wastes tokens a lot.

@SubarcticRec Haha, I wish it were true, but I think it does have a lot of power and I can automate a lot of things right now.

As a founders, what is your take on OpenClaw right now? 1. Still a complete game changer 2. Overhyped nonsense because nobody's actually printing money with it

@AvinashAila @ziwenxu_ OC works with OAI sub OAuth and can the use gemini cli via google oauth. No api tokens needed. 5.4 pro sub is very generous via OAuth.

@ziwenxu_ They are not working with subscription oAurh. Did it work for you?

Most of you missed this: OpenClaw has a hidden ACP Agent. It lets OpenClaw tap into Claude Code, Codex, OpenCode, Gemini CLI without burning tokens on endless back-and-forth just like running Claude Code natively. Drop this config into your OpenClaw and watch it unlock: "acpx": { "enabled": true, "dispatch": {"enabled": true}, "backend": "acpx", "defaultAgent":"claude", "allowAgents": ["claude", "codex", "opencode"], "maxConcurrentSessions": 8, "config": { "permissionMode": "approve-all", "nonInteractivePermissions": "fail" } }

@SlipperyGem yep, i have now two comfy instals: one to run new fp4 models and other for rest. But nvidia latest things require latest drivers and its always risk to update. Some driver versions starts to crash the computer.

@SubarcticRec Aye indeed. Every time you click update could be disastrous. In fact, ever since the recent update, a lot of sub-graphs from the default workflows don't work anymore for me.

For Nvidia GPU Chads out there. You can give this Comfy node a go for the RTX Video Super Resolution for upscaling. Its much faster than Seed VR2, good for vids, but Seed VR2 is still king of creative upscaling, especially of lower res images. github.com/Comfy-Org/Nvid…

"We ship faster than they can clone." 😎

@DrFriesOfficial @wildmindai Its just a paper

@wildmindai have you tested it?

You thought generating code was crazy? Welcome to the next level of vibecoding. Meta just dropped the blueprint for real autonomous coding agents > AI doesn't just write your code anymore > It *is* the debugger > Step into, step over, set breakpoints > Zero Python runtime actually needed > It simulates the entire execution state > It even time travels backwards to find the inputs Literally a world model for code execution arxiv.org/abs/2603.09951

@lucatac0 Qwen 3.5 35B Q 4 26gb ram.

@SubarcticRec true, which models are you running?

the OpenClaw-to-open-box pipeline has begun

@ashen_one fast vision tasks.

anyone have any luck use the smaller qwen local model to do anything significant? mine just handles heartbeats i think lol