“The Miracle” Premieres Now: I gave this film everything I had. Pushed myself creatively and technically, experimented a ton with new tools + models and at times got stuck, started over and wondered if I could pull it off. Excited + nervous to finally share it with you. Enjoy!

Nice work Josh. Fun Halloween family vid.

Have you Seen? Sundays: Tonights pick is have you seen “Force Majeure”? (2014) by Ruben Ostlund

Been rocking with Hermes since the jump, remember Hermes is free, open source and model agnostic. That is a killer combo. Been telling everyone to try since early early this year. Those who know know.

Hermes has an entire library of built-in automations sitting behind one command: /blueprint There are 16 Blueprints covering things like: → Important mail → Morning briefings → News digests → Price watches → Competitor monitoring → Weekly reviews → Meal planning → Daily learning Instead of starting from a blank prompt and figuring out the Cron yourself, you pick the Blueprint you want and Hermes walks you through the setup one question at a time. You choose the details, schedule, and where the results should go. Hermes turns that into the Cron job. If you already know the kind of recurring task you want Hermes handling, /blueprint gives you a much faster starting point.

BLACK MIRROR: FINAL REFLECTION Made with MiniMax H3 on @Hailuo_AI Prompt: FORMAT: 15s | 16:9 | continuous cinematic sequence CONCEPT: Dark-fantasy sword duel inside a gigantic floating liquid-obsidian mirror surrounded by a pure white void. Reflections briefly move independently, creating deceptive attacks. CHARACTER LOCK: Obsidian Ronin: exact reference design, horned kabuto, black mane, dark/gold armor, heavy katana. Heavy, controlled, powerful. Inkshade Valkyrie: exact reference design, porcelain face, two-tone hair, silver-black gothic armor, dissolving cloak, twin blades. Fast, aerial, unpredictable. No redesign. STYLE: Luxury dark-fashion aesthetic, glossy black mirror, silver highlights, smoke, cracks, particles, cinematic motion blur. No blood, gore, guns, battlefield, idle poses, teleportation, hard cuts, or freeze frames. 0–3s White void. Black mirror materializes and ripples. Ronin and Valkyrie emerge from opposite reflections and collide immediately. Huge ripple; reflections continue fighting independently. 3–6s Low camera sweep. Ronin heavy slash → Valkyrie ducks and spins → reflections counter from opposite directions → Ronin blocks → Valkyrie X-slash. Cracks spread across mirror. Smoke forms: REAL OR REFLECTION? then shatters. 6–9s Mirror turns vertical. Both run across it. Ronin overhead strike → Valkyrie aerial dodge → lands behind him → Ronin pivots and blocks. Blades lock; mirror bends like liquid glass. Shockwave separates them. 9–12s Mirror explodes into floating fragments containing distorted reflections. Ronin and Valkyrie leap through them, destroying reflections. Fragments align into a giant black blade shape. Smoke forms: BREAK THE IMAGE → Valkyrie slices through it. 12–15s Brief stillness. Fragments rush together. Both charge. Ronin’s diagonal katana strike and Valkyrie’s spinning twin-blade cross attack collide center-frame. Mirror shatters into thousands of particles. Music cuts to silence. Smoke forms: ONLY ONE REMAINS. Their silhouettes fade as the white void consumes everything. MOTION: Constant action, clear impact/consequence, natural hair/cloak/particle physics. CAMERA: Smooth orbit → low sweep → rapid rotation → dramatic final push. AUDIO: Dark electronic-orchestral score, bass, cello, metallic impacts, blade whistles; silence at final collision. FINAL LOOK: Elegant, aggressive, surreal dark-fashion sword cinema.

Meta now lets you connect Muse to your own hardware to put on your displays or a Raspberry Pi. theverge.com/tech/1004330/me…

This world of AI, Local Models and Creativity moves faster than anything I've ever seen. I've spent the entire year working near 7 day weeks and over 14 hour days to keep up, experiment, build, tinker - and it's been so much fun. But this weekend I'm taking a few days to recalibrate finally, touch some grass. Hang with my son, etc. Been doing too little of that and if anyone else in the same bout, take the moment. Burnout can be real. I think we all need a breather sometimes. Back next week with a curious vengeance. Have a great weekend all!

Awesome just snagged mine! thanks for letting me know @mike_darrow will wear my proudly in addition to my hat!

We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.

H3 Max Reference-to-Video now supports first, middle, and last frame control. Set your keyframes, choose when the middle frame appears, and add image or video references to guide the scene. Shape how your shot starts, unfolds, and ends.

FLUX 3 image by @bfl_ai is here Decide where each element goes, with multiple references in one image, and edit specific details to make it yours Now on Magnific

Mcps and connecting your own agents is a huge win for any model/platform or software that adds it. This is the way and so far I have found it incredibly useful whenever I've done it. Excited to try with Invideo now too.

Ben Affleck (Hollywood star & Artists Equity CEO) talks about how he fine-tunes open video models by unfreezing weights and trained only the last cinematic layer so a film crew can hit real production standards. for context, Ben Affleck founded InterPositive in 2022, a 16-person AI shop for film post and Netflix bought it in March 2026 for $587 mn in cash. He needed that model because public video models were trained on his peers' films, and he did not think that was a real business. So InterPositive raised money, shot its own dataset for 8 months on a controlled stage, and used it only as late-stage training. Each new film then trains a private model on its own dailies, so the production keeps the footage and the learning. That is the product Netflix paid $587 million for. ---- From "Bloomberg Live" YouTube channel, (link in comment)

I'm back from vacation and Hermes update should be around 4x faster for everyone

Introducing JEV-27B-VL. We believe it is the world’s first open-weight, near-SOTA multimodal decision model — built with AutoTrust AI’s Blocks of Experts (BoE) recipe. It does not just describe what it sees. It turns visual state into calibrated action probabilities, then selects what to do next. In this demo, it sees Mario’s position, movement, obstacles, and timing — then chooses whether to run, jump, or do both. See → decide → act.

OpenAI just open-sourced the UI layer for MCP plugins. mcp-extensions lets MCP apps become real ChatGPT surfaces: sidebar apps, file viewers/editors, composer mentions + rich forms. Apache 2.0. ~560 stars in 2 days. github.com/openai/mcp-extens…