composer, art director, digital forensics engineer spreading the joy of open source technologies God’s in his heaven—All’s right with the world
St Petersburg, FL
Joined November 2017
- Tweets460
- Following1K
- Followers22.1K
- Likes63.4K
I'm impressed! Super lightweight. Grab it for your phone or tablet.
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
VARIEN retweeted
No fucking way
ITS 4 AGAIN
DIFFERENT BUG
Announcing 3 Critical vulns for hundreds of millions of IoT, Firewalls, OT, and Military devices including: Palo Alto Networks, Comcast, D-Link, HP, Canon, Honeywell, GE, Oracle, Siemens, Rockwell, Digi, Lockheed Martin, Boeing...
The vulns are with GoAhead Web Server which is a backend library used by these devices. There's a patch available from GoAhead but that doesn't mean there's a patch available for the devices yet -- that's up to the manufacturer.
- TLS traffic over DHE ciphers have a 1-byte secret, "04", meaning anyone can decrypt connections to these devices and steal login credentials and session tokens if the client negotiated with DHE ciphers like the latest TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256, which is extremely likely.
- LibreSSL mTLS has been completely bypassed, meaning anyone can connect over mTLS. If you have one of these devices listening on the internet and are expecting mTLS to protect it from attackers, take it offline or patch asap.
- OpenSSL Certificate Revocation Lists (CRLs) are not enforced, meaning if a user device is stolen or compromised, and you revoke that mTLS cert, nothing actually happens. There are a lot of other attacks this can be used for.
These are critical because these types of devices run our world. Water systems, electricity, military, firewalls, critical infrastructure.
foxio.io/blog/choose-your-ba…
A night shift hospital clerk steals a flash drive that proves his company is selling patient medical files to insurers and has until dawn to outrun the company's private security.
When the flash drive slips from his hand in the chase, a young noodle stand worker finds it in a rain puddle and plugs it in, only to hear her mother's health being sold for $11.
My favorite genre of mathslop is not understanding the golden ratio, drawing a spiral over literally any painting, and going “whoa.”
Invoking 2A with regard to open source models
We believe access to capable open-source cyber models is our constitutional right as Americans under the 2nd Amendment.
It is our duty as technologists and patriots to help protect the digital infrastructure that powers the world's largest economy.
As such, we’re serving our cyber models “at-cost”, in US data centers only, with ZDR enabled by default.
We do not plan to make money serving these models.
If you run a company that is considering how to best protect against cybercrime, reach out to me or @atbeme. We’d be happy to advise.
KYC required. Apply below.
they shot her
but the rxknephew collab was so fucking good
we were early as hell man
still thinking about how @varien and I put a human feature on an AI vocalist’s record literal years ago now
will be funny to watch the future happen downstream of all that
Shame she died though
More reference control. Fewer denoising steps. More places to run H3.
The latest community contributions span 8-step reference-conditioned video + audio, 4-step distillation, sparse attention, and reference-token compression - alongside serving and workflow integrations across GPU servers, TPUs, and Apple Silicon.
What matters is what this enables: more affordable experimentation, finer creative control, and more choices about where and how video generation runs. Open models become more useful when the community can improve the methods, adapt the runtime, and build workflows around them.
Credit to @haoailab @yisun0618 @lanfeng491270 @guocheng_qian @Alissonerdx @ddalcu @tazmannner @HaochengXiUCB @huggingface @Jhuaplin, and the PTD, DMAD, SparkH3, TorchTPU and FlashML teams - including Zhengming Yu and zechengtang on GitHub.
11 additions to the H3 ecosystem index. Thank you to everyone building, testing, and sharing.
github.com/MiniMax-AI/awesom…
This is not a prediction. World Models will be the creative AI "thing" in 2027. I know they've been around, but so have Opus and Photoshop — except now Opus can drive apps better than you or I.
Which means this next batch of world models, mixed with frontier LLMs, will mark the beginning of true "worldbuilding". We've gone from prompting images to prompting entire worlds in three years.
Prompt a world. Set up "cameras" inside it. Add your cast and extras. You now have an entire set without gambling on prompt adherence.
The potential for VR/XR is back.
Welcome to the Metaverse.
Our first look at Bytedance's new World Model, and it looks impressive (Subtitled video):
More than just a world model, check out what it can do with Camera Paths in Stephen Chow's King of Comedy.
Text-to-navigable 3D worlds: A single sentence/description generates an explorable virtual environment.
Free exploration: Users can roam or fly through the scene (e.g., on a paper plane) with keyboard/mouse controls.
Real-time interaction and editing: Spawn objects on demand (such as a spoon), interact with them (climb it), and further modify the scene with additional commands.
Cross-scene connectivity: An “any door”/portal feature that links unrelated environments (e.g., walk out of a room into outer space).
Become Unknowable through Unbroker. Thank you @SHL0MS 🤍 sending a swarm out to <redacted> <specific data broker> right now~
i'm open sourcing UNBROKER: a tool that finds where your personal info is exposed by data brokers and files the removals for you
it runs as a skill in Hermes Agent
_________
your data is everywhere; hundreds of brokers publish your name, current and old addresses, phone, email, birthday, even your relatives. anyone can find where you live in about ten seconds
CCPA, CPRA, GDPR, and a growing number of state laws say a broker has to delete your data if you ask. there's just no easy bulk button. every broker has a different process and many make it intentionally difficult to exercise your right to delete
this is the entire business model for companies like DeleteMe, Incogni, EasyOptOuts; they charge you monthly (DeleteMe is $330/yr for a Family plan) to file removals that you can submit yourself for free, and then you're giving a new company the exact data you want to erase
so i built one you just run for free. your data never even has to leave your machine if you run a local model
_________
how it works:
first it builds search vectors from everything: every name, alias, email, phone, and address you've had (brokers might index you under a maiden name or a house you left in 2014, so the naive "current name + current city" approach can miss profiles). then it fans out parallel sub-agents across the broker list, which refreshes from a maintained public source
automation is tiered. when it can handle a broker end to end with your settings, it drives a browser through the opt-out form, sends the email, and opens the confirmation link itself. soft CAPTCHAs clear on their own with a real browser. anything only a human can finish comes back to you as a short list at the end
the email side doesn't need a stored password and can send opt-outs and open verification links through your own logged-in webmail. you can also wire up SMTP, or keep it manual and just send the drafts it writes
it tailors every request to your jurisdiction, filing under the framework that applies where you live: CCPA and CPRA in California, GDPR in the EU and UK, a general right-to-delete request everywhere else. if you're in California it also uses the state's DROP portal, a single request that covers 500+ registered brokers at once
it holds as little of your data as it can, and keeps it local. dossiers are encrypted at rest if you want, opaque ids keep your real name out of every filename and log, and nothing leaves your machine unless you opt in
brokers sometimes relist you eventually or new ones find your data, so every case is tracked in a ledger and can be re-scanned on a cron schedule so if your data pops back up it files the removal again
github.com/NousResearch/herm…
I'll do better. Prompts below, but most of what you see is after a few passes in post-production software.
Core model:
MiniMax H3 (prefer full or lightly accelerated, not the most aggressive 4-step students if quality is the priority)
Core LoRAs:
fal Realism People Trigger: r34l1sm Strength: 0.7–1.0 Effect: natural skin, better faces/hands, film lighting, less plastic look
MMH3 Realism Slider Strength: 1.0–1.3 Effect: reduces oily/shiny skin, keeps pores and texture
Optional:
Cinematic LoRA (e.g. Cinematic V2) Strength: ~0.7–0.85 Effect: better lighting, color, depth, less AI sheen
Alissonerdx LMS / Sharpness Strength: moderate Effect: extra detail (best with a guide video)
Facial Realism CloseUp Strength: moderate Effect: tighter face detail
Jojocodex Spatial/Physics Strength: 0.3–0.5 Effect: better physics and collisions
4. H3 MINIMAX PROMPT #2 (15s):
Cinematic live-action scene, 15 seconds, 16:9, native synchronized multilingual audio.
REFERENCES: Image 1 is RAN, the auburn-haired woman in the leather jacket with silver ear piercings. Image 2 is SUZU, the long black-haired woman with blunt bangs, facial piercings, visible tattoos, and black tank top. Image 3 is the private Tokyo karaoke room.
Continue the same conversation, preserving both women's exact identities, clothing, voices, seating positions, lighting, and the physical karaoke room.
RAN sits camera-left, SUZU camera-right. The handheld mirror rests face-down on the table. It is no longer being used.
Natural Japanese indie-film cinematography, intimate close-ups, a 35mm lens perspective, subdued neon practical lighting, realistic skin, soft shadows and subtle film grain.
0:00–0:04 — SUZU'S REVEAL
Medium close-up on SUZU'S ACTUAL FACE.
SUZU (English, quick, effortless, matter-of-fact):
"16 gigabytes of unified RAM. That's it. 30 seconds to render."
Pronounce all numbers naturally: "sixteen gigabytes" and "thirty seconds." Give "That's it" a brief confident pause. Dry, understated humor.
0:04–0:10 — RAN DISCOVERS MANDARIN
HARD CUT to RAN'S ACTUAL FACE.
She begins speaking Mandarin naturally, initially unaware she has switched languages. Her eyes widen midway through the sentence as she recognizes what is happening.
RAN (fluent Mandarin Chinese):
"台词表现也不错……哇！等等，我现在是在说中文吗？"
Meaning: "Line delivery is good too... WHOA. Wait, am I speaking Chinese?"
Deliver the opening as relaxed admiration, with correct Mandarin tones and natural connected phrasing.
"哇" is an involuntary surprised exclamation. A tiny beat of realization follows.
"等等" comes more quickly. The final question rises with spontaneous disbelief.
Maintain RAN's established slightly husky, lively female vocal identity. Native Mandarin pronunciation, not English-accented Mandarin. Accurate lip synchronization.
0:10–0:13 — SUZU SPEAKS RUSSIAN
HARD CUT to SUZU'S ACTUAL FACE.
SUZU gives RAN a knowing half-smile and responds in fluent Russian:
"На любом языке. Добро пожаловать в будущее."
Meaning: "Any language. Welcome to the future."
Natural Russian connected speech, correct word stress and vowel reduction, confident understated delivery. A slightly amused inflection, without caricature.
Maintain SUZU's established softer, lower female vocal identity.
0:13–0:15 — RAN'S PORTUGUESE PUNCHLINE
HARD CUT to RAN'S ACTUAL FACE.
She smiles at SUZU, genuinely impressed.
RAN (fluent Brazilian Portuguese):
"Agora isso sim é impressionante."
Meaning: "Now that is impressive."
Natural Brazilian cadence and pronunciation. Emphasize "sim" and "impressionante." Warm, impressed, slightly amused delivery. Maintain Ran's original vocal identity.
End on a shared smile and cut to black.
SOUND DESIGN
Throughout, muffled, thumping Japanese karaoke dance-pop is audible from the room next door. Low-frequency kick drum, indistinct singing, distant laughter, enclosed room acoustics. Continuous across the scene, always quieter than dialogue.
Distinct female voices remain consistent across languages. Natural breathing, expressive timing, accurate multilingual lip synchronization.
FINAL DIRECTION
Every line belongs exclusively to its named speaker. Show the actual speaker's face during dialogue. No talking reflections, no swapped voices, no English substitutions, no subtitles, no narrator, no facial morphing, no cartoonish reactions, no excessive camera movement.
Make the multilingual reveal feel like an effortless, funny moment between two real friends rather than a staged demonstration.
5. POST-PRODUCTION AND FINAL:
Whisper large-v3: transcription, language detection and translation (English, Mandarin, Russian, Brazilian Portuguese)
DaVinci Resolve: node-based color grade (balance, contrast, split tone, vignette
After Effects: Lumetri finishing grade, halation, film grain, and subtitles