Musushi (@0xMusushi) header
Musushi (@0xMusushi) profile picture

Retired Gigablaze on @fogo

Fogo Chain
Joined September 2024
Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)Photo by Musushi (@0xMusushi)
Musushi retweeted
Claude Fable 5.2 + TradingView... just learned to spot trapped buyers & it's f*cking BROKEN this FREE TradingView indicator reads the order flow inside every 5m candle & sends it to AI agents as JSON, & Fable 5.2 decides TAKE, WAIT, SKIP or MANAGE on its own i ported it to python & replayed a real BTC session candle by candle: 43 closed candles, Fable said WAIT 42 times & took ONE short, the moment all 4 steps lined up first understand how this indicator works: > you see candles, an AI agent can't see who's trapped > the indicator reads the order flow inside every 5m candle from its 1m candles & walks 4 steps: environment, location, confirmation, management > it waits for absorption → first shift → second failure, then sends the agent a JSON packet with the entry, the stop beyond the failure & the target how to use it with Fable 5.2: 1. paste order-flow-strategy.pine from the repo into the Pine Editor on TradingView & add it to a 5m chart 2. set the trade window: New York open by default, "Always (24/7)" for crypto 3. the ORDER FLOW STRATEGY panel tracks all 4 steps live on your chart 4. it marks absorption & trapped traders right on the candles, then fires the entry with the stop beyond the failure 5. risk rules are built in: 2 trades a day max & it stops after 2 losses 6. create ONE alert with "any alert() function call" & paste your webhook URL 7. run examples/webhook_receiver.py from the repo & open a free tunnel to it 8. every packet lands in packets/latest.json the moment the candle closes 9. open Claude Code with Fable 5.2, give it agent/AGENT_PROMPT.md & it answers every packet with TAKE, WAIT, SKIP or MANAGE plus exactly why the trade could fail it's FREE & open source, the repo is in the first reply:
these 10 GitHub repos are the most DANGEROUS thing in trading right now together they give you a full AI hedge fund: a research desk, an investment committee and an execution desk in this article I've shown EXACTLY how to build it from scratch
8
22
191
13,968
気になるインジケーター見つけたw 海外ではトレーダーという職業が消えるんじゃないか?って騒がれてる。 Claude Fable 5.2×『tradingview-indicator-for-ai-agents』 無料のTradingViewを使って、 チャートのローソク足をAIが読めるデータに変換して、AIに相場判断までさせられる。 それをチャートに組み込むと、 AIがリアルタイムでローソク足を監視して、条件が揃った時だけシグナルを送ってくる。 仕組みはかなりシンプル。 人間はチャートを見て、 「長い上ヒゲ」 「高値更新」 「出来高増加」 みたいに判断できる。 でもAIは、 TradingViewのチャートをそのまま見ても扱いにくい。 そこでこのインジケーターが、 ローソク足の情報をJSONというAIが扱いやすいデータに変換する。 例えば、 ・始値 ・高値 ・安値 ・終値 ・実体の大きさ ・上ヒゲ、下ヒゲ ・ローソク足パターン ・トレンド ・モメンタム ・出来高 ・相場構造 ・前日高値、安値 など。 そのデータをClaude Fable 5.2に渡して、 ロング / ショート / 静観 を判断させる。 使い方はこんな感じ↓ TradingViewを開く リポジトリにあるコードを Pine Editorへ貼り付けて、 「チャートに追加」 を押す。 株、FX、仮想通貨など、 基本的に色んな銘柄や時間足で使える。 ローソク足を自動で分析 インジケーターが各ローソク足を分解して、 ・実体 ・ヒゲ ・パターン ・トレンド ・出来高 ・主要価格帯 などを数値化する。 ロング / ショートを点数化 セットアップを複数の条件から評価して、 ロングまたはショートを 0〜100点でスコアリング。 一定以上のスコアになった時だけ シグナルを出す。 判定は確定したローソク足を使う。 エントリー・損切り・利確まで表示 シグナルが出ると、 ・エントリー価格 ・損切り ・利確① ・利確② までチャート上に表示。 ボラティリティも考慮して 損切り幅を決める仕組みになっている。 TradingViewからAIへデータ送信 TradingViewでアラートを1つ作り、 Webhook URLを設定する。 するとローソク足が確定するたびに、 分析用データが AI側へ自動送信される。 Claude Codeでデータを受け取る Claude Code側で リポジトリの受信プログラムを動かす。 TradingViewからデータが届いた瞬間に AIが受け取れる状態にする。 AIに相場を判断させる 専用プロンプトをClaudeに読み込ませる。 するとAIが、 ・直近5本のローソク足 ・現在の相場環境 ・上位足の方向 ・トレンド ・リスク などを確認して判断する。 最終的にAIが出すのはこれ ロング ショート 静観 のどれか。 さらに、 ・エントリー ・損切り ・利確 ・ポジションサイズ ・このトレードが失敗する可能性 まで出してくれる。 最初は必ずペーパートレード いきなりリアル資金では使わない。 まずデモやペーパートレードで検証する。 問題なければLINEと接続して、 条件の良いセットアップが出た瞬間にスマホへ通知 みたいな運用もできる。 つまりこれ、 TradingView ↓ ローソク足をデータ化 ↓ Claudeへ送信 ↓ AIが分析 ↓ ロング / ショート / 静観を判断 ↓ スマホへ通知 ここまで自動化する仕組み。
54
413
22
3,549
324,483
Musushi retweeted
vibecoders thank me later. inspora.design
44
162
13
2,454
463,984
Musushi retweeted
A NEW COMFYUI NODE JUST CUT MINIMAX H3 CHARACTER VIDEO FROM 11 MINUTES TO 72 SECONDS Open source. One node. Full 5-second 24fps character video on a single RTX 4090. → 72s warm / 95s first run (was ~650s native) → 13.5 GB Peak VRAM (down from ~22.5 GB) → 4-step Euler sampling with 75% video sparsity → Zero face drift - reference tokens stay dense → ~9x faster than native H3, 2.2x faster than VDN Reference image in, character video out. Identity stays locked. Runs on 16–24 GB GPUs. Repository link below.👇
9
35
4
528
28,754
Musushi retweeted
Another example.. MiniMax H3 vs Seedance 2.5 compared with the original reference. Ran MiniMax H3 on my RTX5090 in @ComfyUI
This is INSANE. MiniMax H3 vs Seedance 2.5 vs Wan 3.0, the quality is near for 100% less. Open-source is getting there! Ran MiniMax H3 locally in ComfyUI.
44
54
6
859
118,861
imagine paying $2k every month for UGC & influencers while this exists. wan 3.0 prime. $0.20 for a 10s video. trust me drop your overpaid influencers.
28
64
3
844
367,897
Musushi retweeted
Hay un agente de trading personal que ya tiene 37 mil estrellas y se instala en un comando. pip install vibe-trading-ai Después lo conectás a Claude Code y tenés un MCP con más de 80 herramientas. Puede hacer research, backtesting, paper trading y manejar crypto (Binance, OKX, CCXT, perps). Lo fuerte no es que “opera solo”. Es que armás equipos de agentes (quant, riesgo, research) y ellos mismos se encargan de probar ideas, revisar resultados y dejar reportes. Vos solo describís lo que querés investigar. Corre local, es open source y no necesita que le des las llaves de tu broker para empezar a experimentar.
16
59
502
29,011
R.I.P. Higgsfield Katana. I reverse engineered it and made a Claude motion design skill that replaces it. Download the skill: scenario.com/explorations/ki… > describe your product, game or brand in a few lines > Claude writes, animates and cuts the whole film to the beat > Scenario makes the music and sound effects, 4K out Higgsfield influencers on Instagram kept promising the Katana setup to anyone who commented. I commented on every one and only got a link to their landing page. So I built it myself, and here it is.
77
205
9
2,514
133,432
Made with Seedance 2.5 on @makeugc (native 15-second video, no editing) PROMPT ⬇️ **CAMERA:** DV 16mm tape camcorder handheld feel. MILA films the opening and closing selfies herself; her gym friend films the dumbbell exercises handheld. Hand shake, misaligned framing, delayed focus pulls, clumsy zooms, occasional face cut-off framing, imperfect shots. Camcorder and camera operator never appear on screen. **LOOK:** Soft, slightly blurry tape quality, faint tape noise, bloomed highlights under gym lighting, flickering auto-exposure, muted contrast, realistic skin tones. **STYLE:** Playful, self-deprecating gym-vlog tone — genuine effort mixed with humor. Quick, casual cuts between excerpts from an ongoing workout. Exercise shots begin and end mid-set, with continuous rhythmic repetitions at normal speed. She talks while moving, without stopping to pose. Time passes between clips; equipment setup and camera handoffs happen off-screen. **Character** MILA — a beautiful American Instagram-style brunette model and performer, 20. Long dark-brown hair in a high ponytail, natural skin texture with a light sweat sheen, glossy lips, expressive eyes, soft glam makeup and delicate sparkly stud earrings. Slim athletic figure with a very large, full bust, very full rounded buttocks, a narrow waist, flat stomach, lean arms and slender legs. She wears a cobalt-blue supportive sports bra with thin straps and a deep scoop neckline showing cleavage, paired with matching ultra-short fitted athletic shorts. Bare shoulders, arms and midriff. White ankle socks, sneakers off. She speaks with an American accent. **Setting** A gym dumbbell area in the evening — weight racks, benches, soft overhead lighting. Two other gym-goers exercise casually in the background, away from MILA. An angled mirror naturally reflects her back, blue shorts and legs in wider shots. She uses the same matching pair of small black hex dumbbells throughout. **Storyboard (15s, 6 shots)** 1. *(0–2.5s, moving arm’s-length selfie)* Already walking past the weight rack, MILA smiles toward the lens and briefly glances at the equipment. Her face and blue sports bra are visible; the framing bobs with her footsteps. MILA: “Arm day. This should be quick.” 2. *(2.5–5s, handheld medium shot)* Cut into an ongoing set of standing alternating dumbbell curls. One arm rises as the other lowers; she keeps the rhythm while speaking. The camera drifts slightly with the operator’s movement. MILA: “These felt lighter yesterday.” 3. *(5–7.5s, handheld wider three-quarter shot)* Another excerpt: MILA is already doing lateral raises, repeatedly lifting and lowering both dumbbells without holding a pose. Her rear view is visible in the mirror. MILA (breathless): “I picked the light ones!” 4. *(7.5–9.5s, handheld close-up insert)* Her hand and forearm grip a dumbbell through the continuing repetitions. The camera briefly loses focus as the weight moves, then catches it again. No dialogue — breathing and distant gym sounds. 5. *(9.5–12s, handheld medium-wide shot)* MILA walks briskly past the camera toward the rack, carrying both dumbbells beside her thighs. The operator turns to follow her as she laughs. MILA: “Yep — my arms just quit.” 6. *(12–15s, arm’s-length selfie after the set)* A clear jump cut: the dumbbells have been returned, and MILA is already holding the camera herself. Walking out of the weights area, she shakes out her free hand and grins. MILA: “Same time tomorrow? No promises.” #AIVideo
🤖 Made with AI
Made with Seedance 2.5 on @makeugc (native 15-second video, no editing) PROMPT ⬇️ **CAMERA:** DV 16mm tape camcorder handheld feel. POV of MILA holding the camera herself, occasionally propping it on the floor or a mat for hands-free core shots. Hand shake, misaligned framing, delayed focus pulls, clumsy zooms, occasional face cut-off framing, imperfect shots. Camcorder never appears on screen. **LOOK:** Soft, slightly blurry tape quality, faint tape noise, bloomed highlights under gym lighting, flickering auto-exposure, muted contrast, realistic skin tones. **STYLE:** Playful, self-deprecating gym-vlog tone — genuine strain mixed with humor, complaining about core work while still pushing through it. Quick handheld cuts, energy staying light and funny throughout rather than fully exhausted. **Character** MILA — a beautiful American Instagram-style brunette model and performer, 20. Long dark-brown hair in a high ponytail, glowing skin with a light sweat sheen, glossy lips, expressive eyes, soft glam makeup and delicate sparkly stud earrings. Slim athletic figure with a very large, full bust, very full rounded buttocks, a narrow waist, flat stomach and lean arms and legs. She wears a cobalt-blue supportive sports bra with thin straps and a deep scoop neckline showing cleavage, paired with matching ultra-short fitted athletic shorts. Bare shoulders, arms and midriff. White ankle socks, sneakers off. She speaks with an American accent. **Setting** A gym mat area in the evening — mirror wall nearby, water bottle on the floor, soft overhead lighting, other equipment visible in the background. **Storyboard (15s, 6 shots)** 1. *(~2.5s, propped camera, medium shot)* She lies down on the mat, propping up on her elbows, already sighing dramatically. MILA: “Core day. Pray for me.” 2. *(~2.5s, propped camera, plank position)* She holds a plank, arms shaking slightly, talking through gritted teeth. MILA (strained): “Why am I shaking already?” 3. *(~2.5s, medium propped shot, crunches)* She moves into a set of crunches, breathing hard, complaining between reps. MILA: “This looked way easier online.” 4. *(~2s, macro insert, shallow DOF)* Close-up on her hands gripping the mat edges during a leg raise, abs visibly working. No dialogue — ambient gym sound only. 5. *(~2.5s, handheld, leg raises)* She finishes a slow leg raise set, flopping back onto the mat immediately after, laughing at herself. MILA (breathless): “Yep—my abs just quit.” 6. *(~3s, arm’s-length selfie finish)* Still lying on the mat, she lifts the camera above her face (camera no visible), grinning tiredly. MILA: “Same time tomorrow? No promises.” #AIVideo
27
81
6
1,062
228,705
Musushi retweeted
the marketing gold rush is here creating high quality animations with claude is really cheap but it’s not easy, nor fast i built an automated workflow skill to make beautiful animations with claude you give it a verse, a quote or a story, and say how it should look it draws every frame, narrates it and scores it → describe the look in your own words: oil paint, watercolour, crayon, ink, storybook → writes every scene as code and paints each frame like hand animation → reads it aloud with a voice that runs on your computer, or use your own → captions timed to every word → music composed for each short, nothing to license → vertical, square or wide → a 40 second short rendered in under a minute on the pro plan you can make 50+ a day, on the 5X you can make 250 i'm giving the whole workflow + skill away for free comment "animator" and i'll send it over
1,495
315
10
4,055
241,959
Musushi retweeted
卧槽!兄弟姐妹们,想用 Opus 5.5 做视频,一定要收藏这 8 个 GitHub 仓库。 1. awesome-opus5-5-videos(约 7100 star) 目前最大的爆款视频合集,每条都附上了背后的提示词。 github.com/yihui-dev/awesome… 2. lemo-opuscar(约 4400 star) 一个 Claude Code skill,装上之后挑一种风格,就能做出纯代码短片,51 种风格随便选。 github.com/lemomo-ai/lemo-op… 3. pdoom-video(约 3000 star) 一支歌词卡拉 OK MV 的完整工程,本地就能渲出 4K60。 github.com/mexicat/pdoom-vid… 4. PDoomVideo(约 2300 star) 同一首歌的另一支 MV 源码,和第 3 个对照着拆,很有意思。 github.com/JohnHeibel/PDoomV… 5. mg-styles-15(约 1400 star) 15 种 MG 动效风格,成片、提示词、源码全都给。 github.com/Vincentwei1021/mg… 6. ClaudeAnimationBase(约 1300 star) 手绘卡通角色动画的起步包,自带 31 种表情。 github.com/JohnHeibel/Claude… 7. awesome-claude-5-5-videos(约 560 star) 视频工作流合集,每个案例都标了出处,Opus 的就有 301 个。 github.com/athemeroy/awesome… 8. opus-video-prompts(约 510 star) 中文案例,配上能直接复制的提示词。 github.com/joeseesun/opus-vi…
143
1,121
83
9,247
850,259
Musushi retweeted
today, Krea 3 makes it to the top of image arenas. - "Fast" pushes the pareto curve of speed and quality. - "Ultra" ranks top-4, only surpassed by OpenAI models. the only company valued under $1T in the top-10! more below 👇
124
190
165
1,634
1,379,628
CLAUDE CREDITS - Redeem it before 7th October Link: claude.ai/code/claim-credit/…
110
321
21
4,877
1,155,053
何億回も言いますが… ChatGPTで作った画像はAI臭が 臭すぎてそのままじゃ使えません。 でもこのプロンプトを一言を足すだけで 別物になります。 特に⑨つ目は 知ってるかどうかで仕上がりが全然違う↓
62
657
24
5,564
972,317
Musushi retweeted
THE CITADEL. HIGH STAKES GUNSHIP BROWSER MMO Closed beta starts this weekend. Claim your spot now thecitadel.com
2,763
2,264
159
3,137
262,470