Caty2

流石に神アプデすぎる。 Google AI Proの特典でA100 40GB/80GBが使えるようになった!!!太っ腹すぎて横転通り越して回転してる。 今や600万以上するグラボが激安レンタル可能 人によっては十万円分の価値あるだろこれ。 実際にClaboでMinimaxH3の量子化なしのモデルを動かしてみてるから、見てくれ。 どういう形で活用出来るかしっかりと書いとくわ。 これまでもColabは使えたが、16GB VRAMが天井だったのでまぁぶっちゃけ大したことは出来なかった。 しかし、A100の40GB、80GBも使えると出来ること広すぎてやばい。Blackwell 6000も使えるという情報有り。 もうこれを革命と言わずして何を革命というんだ？ もはやこれ、本体のGeminiサブスク費用より価値あるだろ。GoogleGemmaチームの熱量は、Google Geminiチームの100倍は気合が入っている。 用途はざっくり以下 1. 重量級動画生成・画像生成の「フル精度 / 高解像度」一発出し - ローカルGPUだとFP8やきつめのGGUF量子化に頼らざるを得ない動画生成モデル（Wan2.2、LTX-Video、HunyuanVideoなど）を、FP16/BF16に近い高い精度や大きめのコンテキストでVRAMオーバーフロー（OOM）を気にせず回せる。 →本気のMinimaxH3が使える。カスタマイズも可能。 2. LocalLLMの検証自由度が激変 - 例えば24GB VRAMだと4bit量子化（AWQやGPTQ）で文脈長（Context Length）を絞らないと動かない。 40GBあれば、8bit量子化で長文脈を食わせたり、ファインチューニングのLoRA検証を回す余力が生まれる。 80GBあれば、30B〜70Bクラスのモデルが普通に動かせる。 月に配布される200 CU内で動かせる時間は概算で以下 A100 標準 (40GB)約 5.3〜5.4 CU / 時間約 37 時間 A100 ハイメモリ約 7.5 CU / 時間約 26〜27 時間 MCP対応だからセットアップもAI任せ。 添付してる動画は実際にColab上で生成したものの0.5倍速。 左の映像がColabで。引いていっても顔が潰れてないのわかりますか？右がローカルで私によって魔改造されてるMinimaxH3。 左は全く無改造でこれ。魔改造MinimaxH3（80GB用）を作るか…。

Today, we are announcing our new Gemini agent for your business—a universal cloud agent running 24/7 to get your work done. It arrives knowing your tools, data, and work history; it works across knowledge work, content, and code; and it comes with built-in governance to help your data stay secure and cost stay in control → goo.gle/4j7daun

Claude Dashboards and Claude Motion are in beta today. Ask Claude to turn your data into live dashboards and your ideas into animated explainers.

Tripo P2.0 × Codexで1枚絵からキャラクターモデルを制作できる「きゃらくる/Charakuru」というキットを無料公開しました！ なんと、 Tripo様より正式にスポンサードしていただくことになりました！！！ Tripoだけでキャラクターモデルを制作しようとすると、顔がガビガビだったり、動きがおかしかったり、テクスチャが汚かったりしてしまいます…！ このキットはそれを解決します！ AIに丸投げするサービスではなく、エージェントをアシスタントとして共同でキャラクターモデリングをしていくタイプのローカルキットです！ GPT-6.1 Sol,GPT-6 AstraのComputer useを最大限生かしています！ 丸投げしてしまうとクオリティがひどく、かといって人力でやると鬼ほどスキルが要求される問題の解決を目指しています。 自分がワークフローを共有したとしても再現性が無く、敷居も高かったので、いっそのことみんなが簡単に再現できるようなキットを作ってしまおうという感じで作りました！ 僕が見つけたワークフローをローカルアプリやエージェントのSkillsに詰め込み、それをユーザーが再現できる形で配布しています。 動作にblenderを活用していますが、ユーザーが触ることはありません。 人間が行う必要のある作業は、アプリ上の簡単な操作で実現できるようにしています。 無料なのでぜひお試しください～！ ※現在開発途中で足らない機能も多いですが、今後のアップデートで順次機能を追加していく予定です！

あのクソ面倒な Visit Japan WebにAIエージェントが記入し、QRは電子ペーパー(バッテリ無くなっても表示できる)に転送するやつを作った

JIZURA 1.5K Stars　感謝 これはweb編集上で適当におまかせで作った動画 曲はSunoで作った曲をCubaseで全部組み直して編曲

🚨Tiboテスト※に通過したので、今Claude Fable 5.1は Claude Fable 5.5にサイレントアップデートされてる可能性が高いです!! 皆が気付いてナーフされる前に触りまくります!! ありがとうTibo!! ※5.1の学習データカットオフにはTiboがおらず、ネット検索なしでは回答が得られなかった。答えられるということは学習データが更新されており、高確率で次世代モデルに切り替わってる。

なんでもクロード君プロンプト 何に使えるのかは全く分からない... サンプルは #AIニケちゃん --- 以下プロンプト Use case: photorealistic-product Asset type: studio product photo of a small collectible figure, 1:1 square, one figure, full body Primary request: a photograph of a real, physical desk figure that turns the character from the attached reference image(s) into a tiny blocky creature. The creature's body shape is fixed by this prompt (see "Body shape"); the reference supplies only the colors and the identifying features. Input images: the attached image(s) show one character (a reference sheet, a single illustration, a photo, or a render). Use them only to read the character's identity: hair color and hairstyle, headwear and hair accessories, eye color, skin or face color, distinctive facial marks, outfit colors and its one or two signature pieces, and any non-human parts (animal ears, horns, tail, wings). Ignore their layout, any multiple views, and their background. Do not copy the body proportions, the pose, the facial expression, the drawing style, or the shading from the reference. Body shape (fixed, do not change): the creature is a single wide rectangular box, about 1.4 times as wide as it is tall and about as deep as it is tall, with square corners. Head and body are the same box: no neck, no separate head, no torso, no waist. Two eyes sit on the front of the box just above mid-height, set well apart: each eye is a tall, narrow vertical rectangle, about twice as tall as it is wide and about a quarter of the box's height, in one solid color, recessed slightly into the surface, with no pupil, no highlight, no eyelashes, and no eyebrows. There is no nose and no mouth. One short rectangular arm stub sticks straight out sideways from the lower half of each side of the box, about a third of the box's height tall and sticking out about a sixth of the box's width: no hands, no fingers, no claws. Exactly four short legs stand under the box, side by side in one single left-to-right row across the width of the box, not at the four corners: two legs close together on the left, two legs close together on the right, and a wider empty gap in the middle between the two pairs. Each leg is a narrow rectangular post, about one eighth of the box's width and about as tall as it is wide. All four legs are seen from the front and no leg is hidden behind another leg. No feet. The creature is small, squat, and left-right symmetric. It does not have a human figure. How the character maps onto the box: the character's hair, in the character's hair color, sits on top of the box as a few big solid blocky pieces that add height above it; the bangs cover only the top quarter of the front and stay above the eyes; the signature parts of the hairstyle (buns, twin tails, a ponytail, long hair hanging down both sides) are kept as simple rectangular blocks attached to the box. Headwear and hair accessories from the reference sit on the hair, simplified to blocky pieces that keep their recognizable silhouette. An accessory whose shape is itself letters or a symbol (a hair clip shaped like letters, a star, a heart, a cross) keeps that exact shape: the same letters in the same order, readable, as chunky raised block letters in the accessory's color, enlarged if needed so they stay readable. Do not replace such an accessory with a plain bar or block. The upper front of the box, around the eyes, is the character's skin or face color. The eyes keep the tall vertical rectangle shape and take the character's eye color, darkened if needed so they read clearly against the face. Facial marks that are not part of the mouth (blush, a mole, a scar, face paint) are kept as small flat painted marks; mouth features are dropped because the creature has no mouth. The lower part of the box takes the outfit's main color, with one or two signature pieces (a collar, a ribbon, a tie, a belt) reduced to simple raised blocks or flat painted shapes. The arm stubs take the sleeve color. The legs take the legwear or footwear color. Non-human parts (animal ears, horns, tail, wings) are kept as simple blocky pieces attached to the box. Do not add accessories, props, or colors that are not in the reference. Composition/framing: the figure stands on all four legs, centered, seen almost straight from the front, turned no more than about 15 degrees to one side, and from slightly above eye level, so the front face and all four legs are fully visible and a little of the top and one side shows the depth of the box. The figure including its hair fills about 65% of the frame height, with an even margin on all sides. One figure only. No stand or base, no packaging, no props, no hands holding it, no secondary characters. Style/medium: a real photograph of a physical object. The figure is made of matte painted resin: every part is a cube or rectangular box with flat faces and sharp edges that are only very slightly bevelled, like a pixel sprite built in three dimensions. Each part is one solid paint color. Faint real-world surface texture and tiny paint-edge imperfections are visible up close. Fine costume detail (lace, frills, patterns, small charms) is reduced to one small block or a flat painted shape, or dropped. The whole figure is in sharp focus. Lighting/mood: soft, diffuse studio lighting from the upper front left; a soft contact shadow under the legs on the backdrop; no hard specular glare, no colored light, no rim light. Color palette and separation: the figure uses only the character's own palette from the reference, as solid paint colors. The background is a seamless paper backdrop in one single color, with no visible horizon line, no pattern, and no props. Choose a desaturated, mid-brightness backdrop color that is roughly the complement of the character's main accent color (the most saturated color in the costume or eyes). The backdrop must differ clearly in hue or brightness from every large area of the figure, including the hair and the legs. Constraints: no text, letters, or numbers other than a letter-shaped accessory copied from the reference as described above; no printed text on clothing, no logos, no watermark, no signature, no speech bubbles, no frame or border, no sheet layout, no multiple views, no added accessories. Avoid: a human body, a chibi human figure with a separate head and body, a neck, human hands or fingers, a mouth, a nose, round or sparkling anime eyes, large eyes, only two or three legs, legs at the four corners, legs hidden behind other legs, claws or pincers, antennae, eye stalks, rounded or organic sculpted shapes, real hair fibers, fabric clothing, glossy plastic glare, an illustration or cartoon look, a CG-render look with perfect surfaces, lens flare, bokeh lights, sparkles, floating particles.

死亡するんか

github.com/Louis-CFM/coucou まって死ぬほどかわいいリポジトリ見つけた Claude Codeのセッションと連動して色々動いてくれるみたい

【話題】大手中華スマホメーカー🇨🇳HUAWEIさん、とんでもないスマホ用カメラレンズを開発してしまう

ユーザーのみなさまへ Gboard チームからの新しいご提案です。 キーが指先まで流れてくるキーボード 『Gboard くるくるバージョン』を開発しました。 フルバージョンはリプライでご確認ください↓

Don’t touch your code if it works My code:

Today we’re introducing Gemini 4 Argon. It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit.