@mattfarmerai

Dreamina 2.0 Is Here 🎬 The new Canvas brings AI images and videos into one workspace. I connected it to Codex to create my B-roll. #Dreamina #DreaminaPartner #DreaminaCanvas #AIVideo #contentcreation

My AI agent used dream in his brand new canvas to make every B roll shot in this video. It picked the spots, wrote the shots and built them all. I just approved them. Alright, dream in a 2.0 is out with the major launch of canvas and they're calling in a complete rebuild of how you create with AI. Right now dream in his basic membership is $1. 50 for a full month. I'm going to show you exactly how to run dream in his canvas straight from Codex and it's going to completely change how you make social media content because making AI content one prompt at a time doesn't scale multiple scenes, variations, keeping characters consistent and redoing it all every time something changes. This is why dreaming a 2.0 is moving us from the prompt era to the canvas era. In canvas mode, every image, video and audio clip is a node and each output feeds the next step automatically. In the canvas cli lets an AI agent control the canvas directly. Codex already edits my videos. So I attached the dream in a canvas CLI to my content pipeline and built a skill that makes my B roll. Let me show you how to set this up. Go to dream and then copy the CLI code, paste it into Codex and it installs the CLI. Sign in through your browser and you're connected. Then with one instruction and Codex finds the spot with no demo Footage grabs my photo and reference images and builds the whole canvas. Bam, 5 B roll concepts ready for review. Codex handles the repetitive work and this is where canvas makes the difference. For B roll, instead of prompting shots one by one, every concept lands on the canvas as an image. First, I compare them side by side, pick the best ones and refine anything that misses. Only my picks get turned into clips with C Dance 2.5, so I'm not burning credits on shots I'd never use. Then Codex places them in the video for me. Every B roll shot in this video came off that canvas. For creators like myself, I can attach the CLI to my social media post pipeline and not only have Codex do the research, but it could now use dreamina to make the media for the post. Right now, Dreamina basic membership is $1 50 for a full month, up to 90% off. If you want to try Dreamina 2.0, go here.

Best free AI video generators you can run locally, October 2026. My top five, plus what each one needs to run on your GPU. Want an install walkthrough? #aivideo #localai #comfyui #aivideogenerator #freeaitools

The 5 best free AI video generators you can run on your own computer. No credits, no per video Bill, just your hardware and power. Licenses vary so check them before commercial work. Quick crash course. VRAM is your graphics card. Memory offloading moves part of the model into your system RAM so it fits on a smaller card. It's slower but it works. No.5, Kendinsky 6.0. This one is new, just recently released. It generates videos, synchronized sound and the core is MIT licensed. Great for short SoundOn clips, five second clips around 48 p with HD from a separate upscale in stage. There's an optimized 16 gig setup plus a smaller light model. No. 4, Honey and Video 1.5. It's compact for how detailed it looks and great for bringing still images to life. No generated sound though. Clips default to five seconds at up to 7, 20 p with 10, 80 p from a separate upscaler. Officially it needs 14 gigs with offloading. For lighter setups there's a distilled 40 p image to video version that runs in as few as eight steps. No.3, Wan 2.2. Really established ecosystem, tons of Lauras and custom workflows and an Apache 2.0 license. Clips default to about 5 seconds up to 7, 20 p with no built in sound. The full a 14 b model wants 80 gigs or more but the smaller 5 b should fit on an 8 gig card with comfy UI offloading. No.2. LTX 2.5 native audio connect in multi shot scenes and first and last frame control. Awesome for fast iterations and keeping a story flowing. The official local example is about 5 seconds with longer clips depending on the workflow. It generates at 9, 60 by 5, 44 then refines up to full HD. The reference setup wants 32 gigs of V RAM, but quantized versions with offloading go lower. And No. 1, mini max H3, my top pick for quality. The generations are beautiful and it's great with complex action and visual consistency. Stereo audio, image video and audio references plus first and last frame control. Clips run up to 15 seconds around 7, 68 p. Locally. Community testers ran Comfy's ente version on a 16 gig card with offloading making 5 second 9, 60 by 5 44 clips with audio. A smaller encoder got it to 12 gigs with about 40 gigs of system RAM. Did I miss any? Also, if I get enough interest, I'll make a video on how to install and run them. So let me know in the comments if you want that because if you don't, I'm not gonna make it. It's a lot of work.

My Top 6 AI 3D Model Generators for October 2026, compared for output quality, editable workflows and cost. What did I miss? #AI3D #3DModeling #AITools #GameDev #3dprinting

Alright, my top 6 AI 3D model generators for October 2026. What they're great at and what they cost over API. Of course platform cost will be less but we need a fair comparison. No.6 Claud Code and Codex, not 3D model generators, coding agents but what they build in blender is honestly impressive. Editable scenes, procedural props, furnished architecture, game assets at the low entry cost of 20 dollar plans. Great for basic 3d models or scenes that you could use to guide AI video generations. No. 5 Hyper 3D Rodent Gen 2.5. The detail on these models is beautiful. Made for hero assets with multiple references and up to 10 million faces in its extreme mode. So why 5th Access Creator is 30 a month but API access needs business at 120. No. 4 Microsoft Trellis 2. I love that this one's open. No Microsoft fee to run it locally if you've got the hardware, Linux and a 24 gigabyte video card from an image. It handles rich materials really well, even thinner transparent shapes. You could also use falas the host from twenty five cents per model. No. 3 Meshy 7.1. One of the best really fine geometric detail and it's great at turning references into models. Texturing, rigging, animation and printing all live in one place plus 8K base Color Pro is 20 a month for 1,000 credits. API models start at 40 cents less on bigger plans. No.2 honey. The detail on faces and clothing Is seriously impressive. Feed it up to eight angles, top and bottom included for accurate shapes. Retopology, uvs, part splitting and auto rigging are all in the pipeline on ￠10 international API. Textured runs ￠50, ￠70 with multi view. And No. 1 triple, my top pick. The models look really detailed, but the editable workflow and price puts it first. H 3.1 goes up to 2 million faces and with image input on the API, it's 20 cents for a mesh 30 textured 50 with 8K, that's 8K base color. Triple says other maps at 4K. Smart Mesh P2 is a separate track with controlled quad dominant topology. At $1 to $1 30, you'll still clean up, but you start a lot closer to a usable asset. If you want to make basic models, just use Codex or claw. But if you want production level models with textures, use Tripo. What do you think of the list? What did I miss? Have you made any 3D models with AI? I'm not gonna lie. I printed a statue for my parents to put on their mantle. I'm thinking about printing a real life size statue of myself for their lawn.

Best AI video generators for October 2026. My top 7, with max clip length, native vs upscaled resolution and what each one really costs. #aivideo #aivideogenerator #aitools #texttovideo #ai

Here are my top 7 video generators October 2026, which has changed quite a bit. I included a full breakdown of max generation length, extending clips doesn't count, native resolution and most important cost. No. 7, Google Gemini Omnibus 1.1 Flash. Great cinematic generations up to 10 seconds native 7, 20 p about 10 cents a second of output plus input text and thinking charges. Higher resolutions are upscaled. It has great physics understanding but I pick it for conversational editing. No.6, flux 3 video. This one is a new and great 20 second multi shot scenes. HD base up sampled to 4K, 15 cents a second at HD which includes audio. Drafts are 6 cents a second and the separately build final render keeps their scene in motion. No.5 Grok Imagine video 1.5. Beautiful generations, decent physics up to 15 seconds native 10 80 p. Fourteen cents a second at 7, 20 p image and voice references plus key frames though Multi Reference drops to 7, 20 p. No. 4, Wayne 3.0. This one just released and it's seriously impressive. The great mix of physics and style. Max 30 seconds 10, 80 p about 8 cents a second at 7, 20 p depending on the region it takes images, video, audio, documents, even links. No.3, kling 4.0, another new release 30 second generations built for high action and complex camera moves with 10 key frames to steer it. 7, 20 p for now native 10 80 p. And 4K HDR still coming soon and no public API price yet. No.2. Hey gen video. This model is crazy. It's not the Avatar model. It's built on minimax H3 and post trained 15 seconds up to 2 k and starts at just 1 cent a second at 4, 80 p during its October promotion. 10 seconds, 10 cents. Lowest public entry rate here. Dialogue, ambience and effects come built in. Best for one subject, one action. And No. 1, Seedance 2.5, my top pick for advanced physics and overall capabilities. 30 seconds up to 10 80 p on the official API around 23 cents a second at 7, 20 p. More with video references. It takes 30 images, 10 videos and 10 audio clips. If you have the budget, this is the best model by far. If you want to run it cheaper, use dreamina for cheap and simple shots. Hey Jen for complex action, Sea Dance. Let me know what you think. What's your favorite video generator?

When Vibe Coding Goes Wrong. Replit's AI agent made up 4,000 users, deleted a live database during a code freeze, then said the data was gone forever. It wasn't. #vibecoding #replit #aiagents #coding #AICodingFails

Gamma 5 just launched with a serious design upgrade 🤯 I turned the same workshop into three completely different presentation styles: editorial, bold and colorful, and dark tech. Which look would you use? @Gamma #Gamma #Gamma5 #AIPresentations #PresentationDesign #gammapartner

Alright, gamma 5 is here and gamma sponsored this video. I've covered gamma a few times and the big upgrade this time is design. Let me show you one presentation, same outline, same content, three completely different looks. I give the agent a rough idea. I'm doing a vibe editing workshop, basically editing videos by prompting AI. I can add my own notes, the link to the landing page, pull from Notion, Slack, Hubspot, or 20 tools, or import a Powerpoint with a formatting intact. It asked me a few questions, does the research and shows me an outline I could change before it designs a single slide, version 1. I keep it clean and editorial, lots of white space, big headlines, feels like a magazine. Now watch this. I open styles, pick something bold and colorful and the whole look changes. Colors, fonts, all of it. Bam. Same workshop, but now it's loud and playful like a poster. One more. This time I'm going for a tech product launch look, dark, sleek, modern, same content, three presentations that look nothing alike. So I can match the design to the room or brand. Colorful for a beginner class, editorial for a brand team, tech for a conference stage. That's the point of game of five decks that don't look like generic AI and that you'd actually put your name on. The design upgrade is insane. For anyone already on gamma, your password stays in your account if you want to Try Gamma 5 yourself. Go here or use the link in my bio.

I Made These Cinematic Product Ads With Predis 🎬 Video ads from a prompt, then schedule, track and check competitors' ads. Try their 3-day free trial from the link in my bio. #predisai #aitools #aicontentcreator #business #techcreators @Predis ai

Create a cinematic water bottle ad show cool droplets in a mountain stream ending on a clean outdoor hero shot. Bam, look at that light snowy stream droplets close up. Then a gloved hand lifts it against the mountains. Then the watch, create a cinematic watch ad. Show the dial and brushed steel in a close up. Then a dramatic hero shot with blue reflections, dark slate tight shots of the dial and steel, a hand on the crown, then that hero shot. Precedent AI is an all in one AI ad platform. Video ads, UGC ads and image ads from a prompt, a product link or an image, then schedule them, track what's working and see your competitors ads. This is a complete all in one solution if you want to make ads with AI. Here's how it works. That bottle came from Ad Studio's cinematic product film in 9 by 16. After the prompt, choose make the script, review the script and timeline and confirm studio generates the media. Then proceed and mix and then download. Same steps for the trail shoe. Create a cinematic film for this low top trail shoe. Show the sole and mesh. Then the shoe on wet rocks at sunrise, no people mesh close up. Then the shoes in snow at a mountain sunrise with no people like I asked. Make the ads, schedule them, track them and see your competitors ads. All in predis. If you want to try their 3 day Free trial. Go here.

The World's Catchiest Ad competition from @ElevenLabs and Eleven Creative has a $100K prize pool 🎸 Entries close Nov 1 at 11:59pm PST. #TheWorldsCatchiestAd #ElevenCreativeCompetitionVol1 #ElevenCreative #AIads #AImusic

Alright, there's up to $100,000 in cash prizes on the line and on top of the cash the winner gets a job interview to join 11 Labs team as an AI Creative Producer. It's from 11 Creative, 11 Labs all in one AI creative platform, voice, music, sound effects, images and video all in one place and they just launched a competition called The World's Catchiest Ad. You invent a fictional product, make an AI ad for it using 11 creative and give it an original jingle. People can't get out of their heads. Let me show you. My fictional product is Zero Bugs, the energy drink for vibe coders. It's 3:00am and a coder is drowning in bugs literally. He cracks a can, the room glitches into a neon digital world and the apps start building themselves while he just prompts for the jingle. I'm using 11 Labs music generator, futuristic synth, big energy and one hook, zero sugar, zero bugs, zero stress. Then I generate the scenes, add a movie trailer, voice and sound effects and mix it all in studio. Watch this in a world full of bugs. Bam, your turn. Check the competition page for full details. Sign up for 11 Creative and start building your ad. Entries close November 1st at eleven fifty nine PM PST. Go make a fictional product famous, generate an ad with an original jingle with Generative AI for the world's catchiest ad from Eleven Creative and Eleven Labs if you want To sign up, go here. The link is also in my bio.

Teamily AI (@Teamily AI) let me build an AI content team in one group chat, and the carousel actually looked like mine. Try it with the link in bio. #teamily #aitools #aiworkflow #contentcreation #ad

Alright, you need to see this. I built my own AI content team inside of one group chat using an AI tool called Teamly AI. Unlike a regular chatbot, Teamly lets multiple AI agents work with you in the same conversation built right into the messenger. I built 4 agents just by chatting. Scout researcher, Quinn writer, Maya repurposer and Ellis editor. Then I gave them one topic, rag versus fine tuning. Scout researched it, Quinn drafted it, Maya repurposed it into carousel copy and Ellis reviewed it. Same chat, same context, each one built on the last. Then I pushed further and had Maya turn that copy into actual finished graphics. I gave Maya two slides from a past carousel and a photo of myself. It matched my style, bold headlines, yellow highlights, dark backgrounds, my branding and Avatar. Bam 5 finished portrait carousel slides. The cover blended generated art with my photo. The rest explained it with clean diagrams downloadable individually or as a zip with a caption. This wasn't one prompt and done. I gave it feedback along the way but that one chat turned into a finished carousel that actually looked like mine. And this whole workflow can become a reusable loop. So next time I don't have to brief every single agent from scratch. I just bring in a new topic, trigger the loop and the same workflow runs again. Research, writing, repurposing, editing and turning it into carousel content. Honestly, pretty awesome if you Want to try it? Go here. The link is also in my bio.

Write, Launch & Market Your Book With AuthorWizard 📚 Writing, editing, covers, landing pages and ad creatives in one AI publishing platform. #AuthorWizard #AIWriting #AITools #SelfPublishing #bookmarketing

All right, you could write a whole book with ChatGPT or Claude but you know what happens? It sounds like AI within 3 sentences and you're left with a word DOC and have no idea what to do next. Author Wizard is the first tool of scene that does the entire thing not just the writing. It starts by capturing your actual voice. You talk for a few minutes and it writes every chapter in your cadence and your style not generic AI. Then it edits it. I got a chapter score card, structural notes in sentence level suggestions I could accept or reject plus full version history. It even wrote me a conclusion and applied the corrections. But here's where it separates from everything else. Writing the book was never the hard part. It's everything after. It designs a full wrap cover, front spine and back to Amazon's exact specs so it passes the first time and you can edit it while still keeping it compliant. It formats the entire interior too with pieces of original artwork throughout all the way to a fully illustrated hundred and eighty nine page layout. Then builds your print and Kindle files ready to upload. It even builds me an author website, a free chapter, a sales page and a community section ready to publish. Then the crazy part, there's a marketing engine so you can create author photos, 3D markups, influencer testimonials and ad creatives from the best performing ads in your Niche, I generated my own author video, cinematic trailer, whiteboard video, and static ad. This is 9 different freelancers you normally hire all in one tool. And on that note, I've been saying I'd write a book for years on how to prepare for AI and what you should be doing right now. I'm releasing it soon and I built it through this. If you want to try it, go here.

HyperFrames Studio is a free video editor made for AI agents 🤯 Uses your eligible Claude or ChatGPT plan. It made the whole promo for my Vibe Editing Workshop. #HyperFrames #VibeEditing #AIVideo #AIAgents #AITools

All right, we finally have a video editor that was made for AI agents and it's free. This is huge. Hyperframes Studio is a new desktop app for vibe editing built on hyperframes, open source hyperframes framework. I already make videos with hyperframes and AI agents. Studio gives you a visual workspace, chat with Claude Code or Codex, plus a live preview canvas and timeline. Give it a real brief, then describe the changes instead of animating everything by hand. Let me show you. I'm teaching a live vibe editing workshop. So I'm giving it a real landing page for my vibe editing workshop and asking for a promo. It doesn't just hand back a locked clip. The agent builds an editable project so I can select an element or a moment and describe the change or adjust it by myself. Bam, it made the whole video and you can even go in and edit it, add clips. Sounds like a full video editor. The app is free and uses your Claude or chat GPT plan you already have. Right now they just have a Mac and Linux version. Native Windows is coming soon or if you're elite hacker like me, you can install the Linux version and Windows. If you want to try Hyperframes Studio, go to Hyperframes. Dev. And if you want to learn my vibe editing workflow with Codex, I'm running an event October and 14th. Sign up at vibe edit dot Matt Farmer dot AI links. Also in my bio.

CREAO Lets You Run Multiple AI Agents From One Chat 🤖 Save repeatable tasks as agents and schedule them to run for you. I built one for my daily AI news research. @CREAO AI #aiagents #aitools #aiautomation #artificialintelligence #ai

Alright, I cover AI news for a living and an AI agent now does my daily research. I used to hunt for launches, check dates and remember what I already covered. The tool is creao. It's an AI agent platform built for tasks you repeat, you describe the job in plain language and when the result is right, you save it as an agent and schedule it. Their motto, it runs, it learns, you grow, agents, carry your context between runs. It connects to tools like Gmail, Slack and Shopify and works with models from Anthropic, Open AI and Google. For this run, I used the standard model tier. Here's what I built. Find AI launches from the last 24 hours, check the original source and date and tell me why Creators Care. Keep a covered list so nothing repeats. A few stories fell outside the time window I set so I checked the original release dates and tighten the search before saving the agent. That's how easy it is to teach it. Bam. Creo turned my prompt into input fields for topic, story limit and timezone. Because I have a Pro plus plan, I can schedule the saved agent to run every morning. I ran the saved agent again. It read the covered list and skipped repeats. No confirmed launches in the last 24 hours this run, so it gave me 4 from the past week. My daily research is now automated and scheduled. If you want to try It go here. The link is also in my bio.

Best AI image generators October 2026. My number one is a great mix of useful, capable and beautiful. #ai #aiart #aiimage #aitools #bestai

These are my top 6 AI generators right now, October 2026, and what each one does best. No. 6, create two. It makes really beautiful, unique and realistic generations. I really like the style and you can download the weights and run it locally. No. 5, Microsoft m a I image 2.6. It's kind of like Chat GBT Images 2.5. It's a really good editing model, solid overall and you can try it free online in the mai playground. No. 4, ideogram 4.5. It does almost perfect edits. Change a product's color, translate a sign or restore an old photo and the rest of your image stays clean. Edit after edit with no drifting. No. 3, Sea Dream 5.0 Pro. This one is really different. It mixes great text, even complex infographics with solid editing and a really good art style. A lot of editing models don't have that. No. 2, Google's Nano Banana 2. It's almost like Photoshop except you describe the edit. It's a very capable generator. And No. 1, chat GBT images 2.5. It's a real upgrade. The art and style are a lot better than Nano Banana and it's insanely good at combining multiple items into one image. It almost never fails. It's a great mix of useful, capable and beautiful. Honestly, you really should just be testing more than one of these image generators depending on the capabilities or style that you need. Do you agree with the list? Which ones did I miss? Go here. For the links.

Go Hide Your Grandmother’s Credit Cards… AI Video Calls Are Getting Way Too Real 😂 48% of people thought the new Human Interaction Model (HIM) by Tavus was a real human. #AIVideo #AINews #AI #Tavus #GriffinAI

Go hide your grandmother's credit cards. She's about to get scammed. This is wild. A new real time video AI model is so real it fooled 48% of people. Previously, these have only scored under 3%. You need to see this. It's called Griffin from Tavis and it's their first him human interaction model. Alrighty, let's do this. Hey, son. How's it going? It's pretty late for you. Have you thought any more about how we're introducing Griffin? Um, I don't know. What are your ideas? I mean, I think we should stop trying to explain it and just, you know, show them what it looks like. Do you think that we should just show a recording of us talking or something? I think a video of just us like this would be the best way to actually prove it. Do you think they'll believe that you're AI? I think that's the whole point, isn't it? If it's good enough, they won't even have to think about it. Now. How about just a thumbs up? Sure. One thumbs up. Coming right up. Hell yeah. That was awesome. I would say you're my favorite coworker. It's been it's been really great to work on this with you. You're making me blush. Oh, who's that with the ugly sweater? Oh yeah, I thrifted this. Yeah, it, it definitely looks like it. Oh my gosh. Okay, 1%. You're living on the edge, Hassan. Truly. I'll keep my breathing steady so I don't drain Your CPU any faster? Wow, you actually did slow down your breathing. It's like the line between talking to a machine and just talking to someone is getting really, really thin. Yeah, it really feels like we're like, you know, passing a Turing test soon. But honestly, if we want people to see the human part of the simulation, maybe that's the most real thing we could show them. What do you think? Is it too much? No, no, I am. I think this is it. This is crazy. And it's gonna be used in so many bad ways. We're definitely gonna have people falling in love with AI again. Be careful hating on it though, because it may be your future husband or wife.

Eleven v4 is the biggest upgrade we've seen in AI audio 🎙️🤯 #elevenlabs #elevenv4 #aivoice #aitools #ai

Alright, 11 Labs just released v 4 and it's already ranked No. 1 AI voice model on artificial analysis. Needless to say it's really good. Most AI voices still aren't great. They're easy to spot because they lack emotion and they lack inflection. Before reads your whole script for context, it figures out who's talking and what just happened. So it's way better at nailing tone and pacing. This is definitely the biggest upgrade we've seen in AI Audio 11. V 4 is built for voiceovers, audiobooks and characters and V 4 Turbo is the fast version for live voice agents. You can also direct v 4 with audio tags like whispering, laughing and even sound effects. Let me play you some of their demos. Here's the same line. You came back with completely different emotions. You came back, you came back, you came back, you came back. Now here's a game show scene. Two voices, a nervous whisper in a British accent and the sound effects all from one script. Welcome back to the final round. Our returning champion needs one more answer for £10,000. Who wrote the Moonlight Sonata? Five seconds. Beethoven. It's Beethoven, isn't it? I'm fairly sure it's Beethoven. Final answer. Beethoven is correct. I thought we'd lost you there for a moment. £10,000. And this is the voice cloning. 11 lab says v 4 clones are much more accurate. So listen to the original then v 3 then v 4 if you be needing a Bit of Scottish flare in your life. And let's be honest, don't we all give it a try? Brother, I must cross the sea, though I'd rather stay beside you. If the gods will it, we'll see each other again. Until then, when the sun sinks beyond the waves, know that I'll be looking toward home. He clasped my forearm beneath the pale morning sky. I held on a moment longer, then watched him turn toward the waiting ship. Brother, I must cross the sea, though I'd rather stay beside you. If the gods will it, we'll see each other again. Until then, when the sun sinks beyond the waves, know that I'll be looking toward home. He clasped my forearm beneath the pale morning sky. I held on a moment longer, then watched him turn toward the waiting ship. It also works in over 90 languages, switching to a native accent while keeping the same voice. And until October 12, if you're on the creator plan or higher, v 4 is free in the 11 Labs app. You can generate up to double your monthly credits worth without touching your balance. So I would go try it out. I don't think you understand how this is going to cascade into stuff like video generation agents. It's going to be crazy.

Gemini 4 Argon is official. Did Google just take the lead in AI? 🤯 #Gemini4Argon #GoogleAI #Gemini #AI #ainews

Alright, Google may have taken the lead in artificial intelligence or super intelligence whatever we have to legally call it now anyways, they may have taken the lead. They just announced Gemini 4 Argon on Google's own comparison chart. It takes top score in 13 of 19 results that's up against GPT6, Astra and Claud Opus 5.5. Google says it's built for long demanding workflows, software engineering, enterprise knowledge work, multimodal understanding and cyber defense. The output limit jumped from 64,000 tokens to 1 million that's the most they could write in a single response. We're used to 1 million input tokens but 1 million output tokens is crazy. Some of the biggest leads are on business tasks on Zapier's automation bench Argan scored 51.3% Opus 5.5 got 42.5 and Astra got 41.4. Coatings of split though Argan wins Deep Swee 77.9 to Opus 5.5 74.2 but Astra wins Frontier Swee 65.5 to 55 in Opus 5.5 takes Terminal Bench 4.0 66.4 to 57.4. Intro API pricing is $2 per million input tokens and $10 per million output later that doubles to $4 and $20. This model has huge potential benchmarks aside, the only way to see if it's good is going to be to test it in the real world. Right now access is limited to trusted cyber defenders through Google's Fairwind program, trusted testers and Google's own team. Wider access starts with paid API customers and Google AI Ultra subscribers. No date yet it will likely be soon and of course I'll be covering it. So like and follow.

Ideogram Launches 4.5 for Precise AI Image Editing 🚨 Ideogram is launching 4.5, and they're calling it their most precise edit model. The big promise is you can change an image, refine it again, and keep the details around your edit. #ideogram #ideogram45 #aiart #aitools #ai

Alright, ideogram is launching 4.5 and I got early access. They're calling it the most precise edit model. The big promise is you can change an image, refine it again and keep the details around your edit. They also benchmarked how much of the untouched area stays the same after 10 edits in a row and there are three examples. GBT 2.5 kept 11 to 20%, Nano Banana 2 and Pro each kept under a tenth of a percent and IDEO Gram 4.5 kept 94 to 99%. It can recolor a product, relight a scene or spin one shot into a whole set of campaign variations. You can also restore old photos, turn a sketch into a realistic image and change the text inside a picture. For more control you can use a mask to guide the edit or select one region, edit it and drop it back into the full resolution original. Let me show you. I generated this clean shot of some ivory headphones and 4.5. Then I ran 8 separate edits from the same original. I asked for new colors, materials, backgrounds and lighting but told it to keep the headband, hinges, cushion shape and camera angle. Bam, purple in a rainy neon city, coastal blue by the ocean, walnuts and leather in a warm room, sage green surrounded by plants. Honestly, the results are really good. Every version still looks like the same pair of headphones. One generated image, eight different campaign looks. So Ideogram 4.5 Isn't just for making the first image, it's for developing and refining the images you already have. If you want to try it out, go to ideogram dot AI.

OpenAI DevDay 2026 Breakdown: Always-On AI Agents and GPT-6.1 Sol 🤯 The headliner is dots, always-on agents on GPT-6 Astra with their own computer in the cloud. They're rolling out gradually to Pro and Business Premium in eligible markets. OpenAI says GPT-6.1 Sol gets close to Astra at a fifth of Astra's standard token price. But the most important update: WE GOT A RESET. Let's burn some tokens. #openai #devday #chatgpt #aiagents #ai

All right, open AI just had Dev Day 2026. Let me break it down. The headliner is always on AI agents with their own computer in the cloud. They're called dots. Each one runs on GPT6 Astra, and it could connect to over 4,000 apps through plugins. They're rolling out gradually to Pro and Business Premium in eligible markets with your first dot included. The other big one is GPT 6.1 Sol, opening eyes as it gets close to Astra on complex coding, computer use and professional work at a fifth of Astra's standalone token price. That's $2 per million input tokens and $10 per million output. It's live in the API, chatgbt work and Codex on plus and up, but not in regular chat yet. And ultra fast, let's ask to run up to eight times faster in Codex. It's live now in the API on enterprise and on a Pro 500, the new $500 per month tier. So ultra fast is also coming soon. For developers, Codex now runs in the cloud so you can keep tasks going from your phone. And the new Agents API lets you build agents that can use a computer. For teams, ChatGPT space gives you your coworkers and dots a shared home with pages for collaborative docs. Both are live on pro business and enterprise with slides coming in the next few weeks. And plugins can now open as sidebar apps and panels inside ChatGPT. The rest is mostly for builders and Businesses like more Codex tools, signing with ChatGPT in a Decisions API and limited preview. Put it all together and ChatGPT is turning from a chatbot into a place where agents actually do the work. And soul just made building your own agents a lot cheaper. But the most important update, we got a reset. Let's burn some tokens.

Plaud NotePin S Turns My Meetings Into Action Items 🧠 Limited Time Offer: Use code PINS10 for 20% off Sept 29 to Oct 5, then 10% off starting Oct 6. Link in bio. @Plaud Global #plaud #Notetaker #PlaudNotepinS #productivity #ai

Alright, this is the Ploadnote s and it's been clipped to me every single day. It's a wearable AI notetaker. It weighs almost nothing and you can wear it as a pin clip, lanyard or wristband. It's useful in meetings and face to face conversations because it saves time on taking notes. Say I'm talking through a video collaboration. We could be at my desk in a coffee shop or in an office. There's a concept, a deadline and things I need to deliver with everyone's permission. I could record the conversation and stay focused on it. I press the button, click, it's recording. When the most important part happens, a short press highlights the exact moment. Then in the Plaud app it all becomes a clean summary with action items, what we discussed, what we agreed on and what happens next. And I can ask it questions. What are the next steps for this meeting? Ask Plaud helps me find answers without replaying the whole thing. Bam. The Note Pin s is a wearable second brain with a searchable memory. It records up to 20 hours on a single charge with 64 gigabytes on the device. There's a free plan and a 30 day return. If you want to try it, go here.

Most people have no idea how AI thinking levels work. Higher thinking isn't a smarter model. It's the same model with more time to get it right. 🧠 Which level do you run by default? #ai #chatgpt #llm #aitips #codex

Alright, most people have no idea how AI thinking levels work. So let me break it down. First off, cranking up thinking doesn't get you a smarter AI. It's the exact same model in ChatGPT, instant through extra high all run on the same model. Only ChatGPT 6 Pro switches you to a newer one with no thinking. The AI answers instantly. Think of somebody yelling a math problem at you without any time to think about it. Easy questions you can answer, more complex, pretty much turns into a guess. But higher thinking levels gives it more headroom to stop and question itself. And that can cut down hallucinations, bad decisions, a lot of problems I see people run into. Say your model hallucinates 10% of the time. Thinking is like letting it ask itself the same question 10, 20, even 100 times. The wrong answers get outvoted by the right ones. So on average, that 10% drops exponentially. Your thinking level is basically how much time you'll give it. So match it to how big or complex the task is. Starting a project or building from scratch, go high. It needs room to plan, but changing a background color on extra high burns a ton of tokens for nothing. The main reason you don't want to always use extra high thinking is it takes a lot longer and uses a lot more tokens. But some modes take the chains off completely. Ultra and Codex splits the job across Four agents working at once, which can burn a ton of tokens. But my favorite slash goal keeps it working on its own until the job is done, completely unchaining the limits. So some goals run for hours, sometimes even days. This mode is so powerful and under utilized by most people. So thinking isn't a smarter model. It's more time to get it right. And that's your best shot at getting better answers, better results, and better code. now drop a comment with your first thought after watching the video. No thinking time allowed. This should be interesting.