json-render now supports YAML as a wire format JSONL... needs a full element before rendering YAML is valid at every prefix, going from element-level to property-level 💨 YAML looks like source code to LLMs And we use 3 standards they know: JSON Patch, Merge Patch, Unified diffshow more

Every AI agent scraping the web right now is... running full chrome to do it. rendering a UI nobody looks at, for a browser that never displays a single pixel. lightpanda skips that entirely. written from scratch in zig, not a chromium fork, and it's 11x faster with 9x less memory than chrome in headless mode. → one-line docker install → drops into playwright, puppeteer, and chromedp through cdp, port 9222 11.8k stars for something that basically asks why you were running a full desktop browser to fetch json.show more

Inviting early testers and contributors to Project Devika -... The open-source alternative to Devin. 👩‍💻 As of now, Devika is far from the capabilities of Devin... but we'll eventually get there. So I am calling the open-source community to join forces! ❤️ Features: - 12 Agentic models that can interact with each other in a feedback loop to understand, browse, research, code, document, and make decisions according to the user's query to complete a project. - Supports Claude 3, GPT-4, GPT-3.5, and Local LLMs via ollama. - Devika can run the code she writes and fix/patch the code herself if she encounters any errors without user intervention. - Devika can deploy static websites she creates on Netlify. (Experimental) - And much more... Will be doing an official launch after intensive testing and bug fixes. 🙌 I've created a Discord server for the early testers and contributors. If you're interested in joining the team, reply to this tweet and I will DM you the invite link. #buildinpublicshow more

Claude Code + Nano Banana 2 is f*cking cracked... 🤯 I built a skill inside Claude Code that writes JSON image prompts for Nano Banana 2, and the outputs look like they came from a professional photo shoot. One plain-text prompt. Claude rewrites it as structured JSON with lighting, camera, composition, style, and negative prompts. Then fires it off to Nano Banana 2. All inside Claude Code. Perfect for DTC brands and agencies who need high-volume ad creative without booking a shoot. If you're using Nano Banana 2 for product shots and lifestyle images but every generation feels like pulling a slot machine lever — random lighting, inconsistent style, plastic skin, misspelled labels ... This skill fixes the entire output: → You describe what you want in plain English → Claude rewrites it as a structured JSON prompt (lighting, camera angle, lens, depth of field, color grading — all of it) → Fires it to Nano Banana 2 via API → Saves the prompt + image in organized folders → You iterate on the style until it's dialed, then every output matches No more slot machine prompting. No more inconsistent brand imagery. No more burning credits on unusable generations. What you get: - Photo-realistic product shots and lifestyle images on demand - Full control over style, lighting, composition, and camera settings - Saved JSON prompts you can reuse across every campaign - A skill that gets smarter the more feedback you give it Built 100% in Claude Code with a custom skill + Python scripts. I put together a full playbook showing the exact skill, the JSON schema, and the workflow to set this up yourself. Want the full playbook? > Like this post > Comment "BANANA" And I'll send it over (must be following so I can DM)show more

Claude Code + Nano Banana 2 is f*cking cracked... 🤯 I built a skill inside Claude Code that writes JSON image prompts for Nano Banana 2, and the outputs look like they came from a professional photo shoot. One plain-text prompt. Claude rewrites it as structured JSON with lighting, camera, composition, style, and negative prompts. Then fires it off to Nano Banana 2. All inside Claude Code. Perfect for DTC brands and agencies who need high-volume ad creative without booking a shoot. If you're using Nano Banana 2 for product shots and lifestyle images but every generation feels like pulling a slot machine lever — random lighting, inconsistent style, plastic skin, misspelled labels ... This skill fixes the entire output: → You describe what you want in plain English → Claude rewrites it as a structured JSON prompt (lighting, camera angle, lens, depth of field, color grading — all of it) → Fires it to Nano Banana 2 via API → Saves the prompt + image in organized folders → You iterate on the style until it's dialed, then every output matches No more slot machine prompting. No more inconsistent brand imagery. No more burning credits on unusable generations. What you get: - Photo-realistic product shots and lifestyle images on demand - Full control over style, lighting, composition, and camera settings - Saved JSON prompts you can reuse across every campaign - A skill that gets smarter the more feedback you give it Built 100% in Claude Code with a custom skill + Python scripts. I put together a full playbook showing the exact skill, the JSON schema, and the workflow to set this up yourself. Want the full playbook? > Like this post > Comment "BANANA" And I'll send it over (must be following so I can DM)show more

Introducing Kaleido💮 from AI at Meta — a universal... generative neural rendering engine for photorealistic, unified object and scene view synthesis. Kaleido is built on a simple but powerful design philosophy: 3D perception is a form of visual common sense. Following this idea, we formulate rendering purely as a sequence-to-sequence generation problem, successfully unifying neural rendering with the architecture principles behind modern language and video models. Unlike traditional neural rendering methods, Kaleido learns 3D purely in a data-driven way, without explicit 3D representations or structures. It acquires spatial understanding directly through large-scale video pretraining, then multi-view 3D data finetuning, inspired by how LLMs acquire textual common sense from large corpora before specialising in domains like coding. Through extensive ablations, we progressively modernised the architecture design and training strategies and tackled key scaling challenges in sequence-to-sequence generative rendering, arriving at a design that’s simple, versatile, and scalable. Kaleido significantly outperforms prior generative models in few-view settings, and remarkably is the first zero-shot generative method matches InstantNGP-level rendering quality in multi-view settings. We view Kaleido also as an alternative step towards world modeling that flexibly spans a spectrum of “realities": with many views, it faithfully reconstructs grounded reality; with fewer views, it imagines plausible unseen details. 🔗 Explore more results and paper:show more

"Hmm, a word I've never seen written before. I... will sound it out based on the letters that compose it. Cah-ah-tuh. Cat! I've heard that word before, and now I know how to spell it! How intuitive!" That's how we used to teach reading. It had a proven track record of success. Now they literally tell kids to just guess what the word is and actively discourage them from sounding them out, and are surprised less than a third of American eighth graders can read at grade level.show more

Heres an actual way to make $10k a month... from TikTok + organic affiliate TT slides is probably the best way to make AI content for a few reasons > easy and not time consuming to make > much harder to detect images are AI > slideshows rarely get the AI label from TT (especially if you do what I’m gonna show you) >slideshows require less engagement to go viral (more consistent virality) >CTA can be more natural this AI slideshow format is going insanely viral consistently on brand new accounts and no one is even detecting it’s AI they’re super easy to make and use a story telling format that’s really smart you can take the exact same formula to promote sweeps offers from Glitchy and make $10k a month pretty easily here’s the blueprint >Content creation to create the slides realistic you can simply take a photo from Pinterest and put it into Gemini or Chat GPT and ask it to give you an EXACT JSON to recreate the image add in any details you want to add like “make her hair blonde” “make her eyes blue” take that JSON and put it into Nana Banana If you want her in certain backgrounds do the same process and add it in with the JSON of the girl you’ve created or just describe it Now clear the meta data from said image to stop getting the AI label If you still get it go to a meta data analysis site and put the data in ChatGPT Ask if their is anything in the meta data that is signalling this to TT >Writing scripts The reason these go so well it’s because they have a negative scroll stopping hook that instantly make you want to know what the slides going to say next “got fired from my job” “failed my exams” Along with the music that sets the emotion of the video It wouldn’t work as well if they had a random viral song that’s up beat plus the cherry on top is that the image correlates with what’s being said in the hook it’s self acting as a visual hook It wouldn’t work aswell if it was just an image of a girl on her bedroom >how to interpret sweeps Choose an sweeps offer from Glitchy for a retail store (Walmart/target) follow the same format of scroll stopping negative hook Example: “broke my arm” then you would tell a story that paints a bad working environment, maybe she got fired for breaking her arm and is now exposing secrets and your CTA would then be “they don’t promote this but they have a secret feedback program” this is just to give you inspiration but there is literally countless ways >account set up - US proxy / US sim - Download TT with US proxy on - Buy aged account (to help with account trust and getting banned due to proxy issues) - warm up for 2 days (scroll vids all the way through, like, comment authentic things relating to video) Then you post This is a very good way to at least reach a couple K a month but I’d be surprised if you don’t reach $10k beyondshow more

We know what a downburst looks like… but this... is Colorado, and it was below freezing at ground level. This was a snow event… a cloud burst of snow. But what else do you notice? A burst from f***ing what? Where’s the big cloud? I watched this come together from nothing in minutes- there WAS NO CLOUD five minutes earlier. This is not how pressure and temperature and vapor work at these temperatures… unless you have a GEC amplification event. I’ve now seen 6 since 2017. They never break the boundary layer but you spot their influence. What happens when these form the column all the way to the ground?show more

We are now back to full blown hard work... after an incredible week at #aptosworldtour We are more bullish on Aptos than ever after seeing the incredible team and developers enjoying the MOVE language. But we are back now and all about #Propbase so we need to drop an update to timings for April as we have a better idea now! Remember these dates are all tentative as we are working overtime to get everything out in April Mega Announcement 3 – 17th of April Initially we wanted to be ready by 12th of April but we need to polish the presentation a little more so we now anticipate the announcement on the 17th of April. We are incredibly bullish on this announcement and can’t overstate how huge this will be for Propbase! This will be the biggest announcement to date! NEXUS – 27th of April We want to prove how incredible Nexus is so we will get our first property listed before our next top 10 CEX listing, this is a mega bull move in our opinion. We can tell from our DM’s of people trying to get their accounts ready to buy property, that this initial property launch won’t last long! CEX Launch date – 30th of April How do we plan to finish the month off? TOP 10 CEX LISTING. So we will drop our Mega announcement and let the community get absolutely pumped over the news then we will see a 1 minute sell out on our first property listing (what we anticipate) and then move to our largest CEX to date. Going to be a blockbuster month! $PROPS #PROPS #Propbase #RWA #APTOS #BlackRock #Propertyshow more

Hi gamers, Here’s some quick additional context on the... patch, split into a few posts Core rules It’s clear that Pushblock needs buffs, so we gave it some. It pushes farther but with 4 more frames of recovery. Specifically, we want champions to be able to pressure after a jump-in or a std.L if they’re on top of it and react to the pushblock in time, but be more reliable on defense. Pushblock and Break now enable you to protect assists by tagging. It won't "cost health" to use pushblock in more scenarios. Freestyle is getting a Break penalty adjustment, going from 35% to 20%. We want there to be a penalty, but it was too severe. You’ll still feel it, just not as much. Hurtboxes are now standardized between champions while standing and crouching. This will be invisible in most situations, but makes spacing-specific combos much more predictable. Wall slump hurtboxes will also no longer be character specific, and now every champion will have the same combo against wall slumped opponents.show more

This is why it’s going to take a VERY... long time for our USMNT to get to be at the level of the world’s best. I mean look at the footwork from Romero of Argentina in a semifinal match for the World Cup. Under this pressure, a CENTRAL DEFENDER pulls off this dribble to get himself out of trouble …pulling it out of his own ass. It will be a long time before we pull off creating a defender of this caliber able to do this with all the weight on their shoulders as they try and play it out of the back.show more

♾ Credit System Integrated + Vision Announcement♾️ NFINITY AI... is going to become a household name in both Web3 and the AI space. We keep repeating the same thing over and over again, and we will do it again today: ♾️“A New Standard” ♾️ Our team is filled with experts. Years and years of experience comes together in this project. Not only do we understand Web3, but we also understand Web2. We analyzed, executed, and tried a lot of different approaches. We made dozens of case studies and came to one conclusion: Web3 needs a new standard. Web3 needs Web2 copywriting. Web3 needs non-greedy developers. Web3 needs communication and an understanding of human psychology at the highest level. Web3 needs speed and patience. Web3 needs timing. KOLs aren’t bad; they just have to be managed correctly. Web3 needs common sense. All that crazy 1000x stuff is not going to bring us great tech. It’s not providing us with the things we need. The short-term minded rule this place, while long-term people should be on top. And that’s what we’re doing. We like to take time to get to know our core community. We want to build with smart people, people who want to invest but also contribute. We want to build with the top 0.1%. The plan is simple: We’re building out an excellent idea. We start with Stage 1 and perfect it with the core community. We build out to Stage 2 and bring it to the masses. And when we arrive at stage 3 and start exploring the possibilities behind an L2... Well, then it’s time to shine. Build together with us. To Infinity And Beyond.show more

Contrail lesson! 1. “Chemtrails” don’t exist. Just to get... that out of the way. 2. Observe the satellite loop and Skew-T chart. In the IR satellite loop you can see yesterday, the West Coast had a decent short wave ridge suppressing moisture over California and Nevada. Today, you can see moisture from a low pressure over the Pacific spilling over the ridge that is now moving east of California. This is upper level moisture ADVECTING into the area. This upper level moisture is mainly above the 500mb level, or 20,000ft. 3. Now observe the Skew-T chart. Particularly clue into the 300mb level. This is a perfect example of what I talk about all the time, and why it’s important to pay attention to the 300mb level. This moisture layer is advecting particularly at the 300mb level, and synoptic scale cirrus development, and advection, typically occurs at 300mb. This is key because aircraft are flying at and above the 300mb level. 4. So, lastly, observe the pictures that I took of the sky over northern Nevada at the time of this post. You can see the layer of cirrus as well as contrails persisting in that moisture layer, exactly as depicted in the satellite shot AND confirmed by the Skew-T chart. Keep in mind that temperatures at this level of the atmosphere are typically -20 to -50°C. In this case, you can see that the temperature at 300mb is -40°C and relative humidities at this level are far different than what you experience at the surface. Any decrease in the gap between temperature and dewpoint at this level can significantly increase the relative humidity. This is why it’s referred to as “relative”because it’s far different than temperatures and dew points at the surface. So, to bring it all together, aircraft flying at these altitudes, which most commercial and military aircraft do, injecting warm, moist air from the engines rapidly into the super cooled environment, not only instantly form contrails, but when relative humidities are as depicted in this example, will enable contrails to persist for hours at a time supported by the moisture existing in that layer. This is what causes persistent contrails. These ARE NOT “chemtrails” and because they persist, does not, and will not ever, make them “chemtrails.” Now that you all needed your government to tell you that climate change was a hoax and I’ve been telling you for years that the “Geoengineering” and “chemtrail” nonsense are propaganda directly related to the climate change hoax, hopefully you can take some time to learn the basics of the atmosphere and understand what I’m showing you here, and how it works, so you’re not fooled by climate propaganda going forward. Thank you for your attention to this matter. 💪🏼🇺🇸show more

OpenClaw, but built for normal people. Sim is an... open-source platform that lets you build AI agent workflows on a drag-and-drop canvas. Connect them to channels like Telegram and WhatsApp and deploy without writing a single line of code. They also have a built-in Copilot that generates entire workflows from plain English, which you can then tweak and customize in the UI. Key features: - Free and open-source (Apache 2.0) - Vector store integration for RAG-grounded agents - Self-host with one command (`npx simstudio`) - Run fully local with Ollama, no API keys needed - Supports vLLM for production-grade self-hosted inference The thing I really like about Sim is the level of control you get. You can add conditional branching, parallel execution, human-in-the-loop approval gates, and even nest workflows inside other workflows. Everything is visible on the canvas, so you know exactly what your agent is doing at every step. And you can build a workflow in Sim, deploy it as an MCP server, and plug it into any agent, including OpenClaw. I've shared the link to Sim's GitHub repo in the next tweet.show more

bitcoin:native UPDATE My local view on #BTC My plan:... 1. We got an impulsive move up from the $62K/$64K support zone (DONE) 2. We hit all the Fibonacci targets (DONE) 3. We stopped at Strong Resistance - $83K (DONE) 4. The key zone is $83K/$85K 5. Scenario one - we get a pullback now, followed by a move up toward the key zone (LOAD) 6. Scenario two - we move straight up into the key zone (X) 7. Long zone 1 - 74к/72к and long zone 2 - 72к/69к This is how I currently see bitcoin:native There are some interesting similarities if we study the 2023 price action. Back then, we got a move up from the 0.23 Fibonacci level, and it was only in March 2023 that we came back to this Fibonacci structure to test the 0.382 level. If we draw a new Fibonacci range, that would correspond to the 0.5 level So my view is that I only see $100K for bitcoin:native next year $ETH and $SOL have already outperformed #BTC, both in percentage terms and on the chart Right now, we have already tested the 0.23 Fibonacci level, but there was no reaction. This means there is a high probability that we move lower, where we could look for long positions targeting $83K+ REMINDER: I warned about the BTC bull trap at $82K, the bounce from $62K, the summer drop, and the SPACE X decline before they even happened. FOLLOW ME AND TURN ON NOTIFICATIONS. MOST PEOPLE WILL FOLLOW ME TOO LATE.show more

🚨 $HYPE is going exactly as I told you.... We saw a correction into the 56–60 zone. NOW: HYPE is heading toward the target $65–$70! We are seeing a bull trap on this asset. After testing the 0.5 Fibonacci level, we got a local reversal pattern: an inverted head and shoulders. New Fibonacci target: 0.5–0.618 = $65–$70 This is just a bounce before the further drop to $38–$48! Spot buys start at $48. Not before. Share your opinion in the comments - what do you expect from this asset? As a reminder, I warned about the bull trap at $82K and the summer drop MY NEXT CALL WILL BE THE BIGGEST ONE OF THIS CYCLE TURN ON NOTIFICATIONS. MOST PEOPLE WILL FOLLOW ME TOO LATEshow more

Qwen 3.8 Flash decisively crushes the M5 Ultra with... this new recipe by vr8vr8 . I think I found the fastest dual DGX Spark recipe for Qwen3.8-Flash-Next. 138 tok/s structured, 133 code, 383 structured aggregate at four streams, and now it talks to Hermes. It feels like a cloud API, running on your desk. This should torch any current M5 run right now. This is part two. Same pair of Sparks, vr8vr8's recipe, his new v4.1, plus a patch of mine that's now an open PR against his repo. WHAT INITIALLY BROKE Through Hermes, the recipe answered in garbled mixed Chinese, then looped on "function86 to90 turn91" until it hit the token cap. Most people would blame the quant. The quant is fine. Three things in the serving stack were stacking up: Hermes doesn't send a temperature for this model. When a request omits it, vLLM 0.30 applies the checkpoint's own generation config: temperature 1.0, top_p 0.95, top_k 20, sampling on. The documented opt-out flag still leaves the server default at 1.0. Every Hermes turn was sampling hot. Hermes sends reasoning as a top-level reasoning object. vLLM 0.30 only maps a field called reasoning_effort onto thinking. So "thinking off" never arrived. Thinking is on by default in the chat template, and with no effort set it defaults to xhigh. A request for no thinking became the longest think the model can do, at temperature 1.0. This is why benchmarks alone do not tell you if a recipe is ready. A real harness that leaves temperature out and names the reasoning field differently gets a different model. Mia's older image never applied the generation config and drafts with argmax, so omitted temperature never sampled hot there. This recipe samples the draft. THE PATCH Two changes to vLLM's chat protocol, applied on both ranks by The Hermes reasoning object now sets thinking. none turns it off. low, medium and xhigh pass through to the template. An explicit enable_thinking still wins. If a request omits temperature, the chat path uses 0. An explicit temperature still passes through. Receipts: three table requests with no temperature and thinking off came back in English. A medium request closed its think and answered in English. The PR is open, not merged. Until it lands, a stock Hermes custom provider still samples at 1.0. HIS V4.1 vr8vr8 shipped recipe v4.1 in the meantime, same v6 image: FlashInfer GDN prefill. His card claims about +5% prefill with decode unchanged. On my depth ladder, prefill came in +6 to +7% from 128k to 900k. Claim holds. THE SECOND GRID Same pair, same protocols, thinking off, temperature 0. My runners already send temperature 0, so this is v4.1 against part one, not a test of the patch. Prose 600: 81.6 tok/s (part one 78.4) Structured count 200: 138.3 (131.8) Code clamp: 133.2 (126.4) JSON catalog: 118.6 (111.8) That's up 4 to 6% across the board, which is run-to-run band. I'm not crediting FlashInfer for it. At four streams: structured 383.1 aggregate, code 330.5, JSON 290.4, prose 200.5. Single-stream structured costs 0.52 J per token of GPU power across both boxes. Tool-call smoke test clean, finish_reason tool_calls, no XML in content. DEPTH Needle retrieval 18/18, from 8k out to the 900k rung (880k server-measured). Decode after fill holds between 65 and 77 tok/s the whole way, 70.6 at the top. Prefill runs 3,245 tok/s at 128k and still 2,577 at 880k. AGENT TURNS The OMP agent lane now runs a full 2,048-token write_file turn at 35k context, replacing part one's 35-token window: 107.4 tok/s decode. The Hermes research turn, two web_search calls at 35k context: 81.0 tok/s, up from 74.7. STILL NOT LIKE-FOR-LIKE Against Mia's d2f54b7 recipe on this same pair: prose 1.6x, structured 2.0x, code 2.0x, JSON 1.9x, Hermes turn 2.2x. Different quant, K=5 vs K=3 drafting, bf16 KV vs fp8. My own version is also my own customized recipe, but it felt better to just serve a PR to the creator, since I didn't invent anything new. The biggest single lever is his output head: 1.18 GiB down to 0.33 GiB, read about 5.4 times every decode step, worth +30% engine steps on its own. SUMMARY Every clean lane faster than the mainstream recipe on this rig: checked. Retrieval 18/18 to 900k: checked. Decode holds to 900k: checked. Agent lane measured over a full 2,048-token window: checked. Hermes out of the box: PR open, work in progress. The speed was always the hard part, and vr8vr8 had it in part one. Sources , Benchmarks and Recipe in Reply 👇show more