Gerald A

Programador autodidacta fascinado por la inteligencia artificial y sus aplicaciones. Siempre buscando nuevos retos y soluciones innovadoras. #AI #ML

[硅基废话 @DDJCXX]

上面这个号的打法，我们昨晚跑通了。

不用拍，不用露脸。

找一条火过的视频，把人换掉，就是一条新的爆款。
两个工具就够了：ChatGPT 做角色，Seedance 2.5 换人。

这是我们造的角色，完整步骤在下面👇

[Gilberto Marcano @gilbmarcano]

@DotCSV Copia barata pero interesante de la canción que está por ahí viral. Tiene mil fallos y no es ni buena, pero poder hacer esto en 1 hora con Suno + Opus 5.5 me parece loquisimo. Imagina invirtiéndole todo un día.

[robert. @robrtcode]

🚨| ASÍ ES COMO JEV Y SEIS GROK BOTS HAN CONVERTIDO $1.000 EN $20.974 EN CUATRO NOCHES.

Un trader ha dejado su mesa entera en manos de siete agentes. Ninguno es una persona.

$1.000 → $3.833 → $9.118 → $12.539 → $20.974

La última noche la mesa hizo un 67%. Solo el último turno sumó $8.435.

Él mismo admite que era el peor trader de esa mesa. Compraba en el techo y se quedaba pillado.

Ahora cada agente tiene un solo trabajo.

→ Jev elige la moneda. Miró 81 y descartó 80. No habla nunca.

→ Tape lee el gráfico.

→ Price espera la caída y entra.

→ Wire revisa que la cuenta de X de la moneda no sea falsa.

→ Size decide cuánto se juega, nunca más de un tercio de la banca.

→ Exec mete la orden.

→ Risk corta todo lo que se tuerce.

Los seis de debajo de Jev son Grok Bots.

Quien compró la moneda al abrir y aguantó se llevó un 26%. Con el mismo gráfico, la mesa sacó un 67%.

Fueron cinco operaciones. La mejor dejó $4.912. La única mala fue Price persiguiendo un rebote, y Risk la cortó a la hora en -$319.

Nadie cobra sueldo. Todas las decisiones de Jev del turno costaron menos de un céntimo.

Se despidió a sí mismo de su propia mesa. Y la mesa va mejor sin él.

Esta noche empieza el quinto día, con los mismos siete.

[atomic.chat @atomic_chat_hq]

Local Laya moggs Jev at @grok 4.7-built Tetris 🧩

An open-weights System One model called Laya, beat cloud-based Jev at playing Tetris by making decisions 11 times faster, running locally on a 16GB MacBook Air!

Run AI models locally -> https://t.co/RbcCOIgVkj

[火山哥🕊️ @huoshan007]

兄弟们，这个开源 AI 视频项目有点吓人。

GitHub 6多万星，不是没原因的。

我把一句“做一个 AI 工具测评短视频”丢进去，它自己拆脚本、找素材、配音、上字幕，最后真给我吐了一版成片。

更狠的是，我又塞了条爆款视频进去，它把对方的节奏、镜头、文案钩子全扒出来，还给了几套仿写方案。

我用它跑了一条 58 秒的视频，成片虽然还有点糙，但结构已经像那么回事了。

这感觉很怪。

你说它是工具吧，它背后 12 条流程、100 多个工具、700 多个 Agent 技能，怎么看都像一个小型剪辑组。

链接在这：
https://t.co/gQmnzc47R6

做短视频的真别只收藏不看。

我试完就一个反应：剪辑师还没下班，AI 已经开始抢夜班了。

[山羊赚钱笔记 @shanyanggm]

卧槽兄弟们，GitHub 上真的有一堆免费到离谱的项目。

很多能力已经能直接干掉你正在月付的软件。

1. TradingAgents
AI 多 Agent 量化交易框架
https://t.co/3T1sPPQ6TL

2. LibreChat
一个界面接入 ChatGPT、Claude、Gemini 等多个模型
https://t.co/HZJQEij5h1

3. HyperFrames HeyGen
开源的视频生成引擎
https://t.co/o0OnS8iJMG

4. Fincept Terminal
开源版彭博终端
https://t.co/n5COGtvHc4

5. MoneyPrinterTurbo
AI 一键生成短视频 https://t.co/Po8sGObI7q

6. Agentic Inbox Cloudflare
开源 AI 邮件助手
https://t.co/S7XDWi1yTD

7. VoxCPM
AI 声音克隆工具
https://t.co/wkSGMchuq2

8. Flowsint
开源 OSINT 情报分析工具 https://t.co/RwFvhurB0P

9. agent-skills
Claude Code 技能库 https://t.co/c2NifSzKM1

10. Nango
开源 API 集成平台
https://t.co/tNbwrMEzZn

兄弟们，这些真不是玩具项目。

很多你还在交月费的软件，GitHub 上已经有人做出开源平替了。

一句话：

别只会收藏 AI 工具网站了。

真正狠的东西，很多都藏在 GitHub 里。

[M. Asif @meAsifAi]

Created with @FlowbyGoogle
Character Sheet Prompt :
Create a professional, ultra-photorealistic CHARACTER CONSISTENCY SHEET using the uploaded reference image as the ONLY source for the character's identity.

IMPORTANT — IDENTITY LOCK:
The uploaded reference image is the definitive identity reference. Reproduce the EXACT SAME PERSON throughout the entire character sheet.

Do NOT redesign, beautify, reinterpret, or replace the character.

Preserve with maximum accuracy:
- exact facial identity
- facial structure and proportions
- eyes, eye shape and iris color
- eyebrows and natural brow shape
- nose shape and proportions
- lips, mouth shape and fullness
- cheeks, jawline and chin
- skin tone and natural skin texture
- age appearance
- hairline
- exact hair color, length, texture and hairstyle
- body proportions and physique
- distinctive facial characteristics

CHARACTER SHEET LAYOUT:

Create one clean, professional reference-board image containing:

1. LARGE FRONT PORTRAIT / FACE CLOSE-UP
2. FRONT VIEW
3. 3/4 LEFT VIEW
4. 3/4 RIGHT VIEW
5. LEFT PROFILE
6. RIGHT PROFILE
7. BACK VIEW
8. FULL-BODY FRONT VIEW
9. FULL-BODY 3/4 VIEW
10. FULL-BODY SIDE VIEW
11. FULL-BODY BACK VIEW
12. NEUTRAL EXPRESSION
13. NATURAL SMILE
14. SOFT SMILE
15. LOOKING LEFT
16. LOOKING RIGHT
17. EYE CLOSE-UP
18. EYEBROW CLOSE-UP
19. NOSE CLOSE-UP
20. LIPS CLOSE-UP
21. HAIR CLOSE-UP
22. ACCESSORIES CLOSE-UP
23. OUTFIT / CLOTHING REFERENCE
24. COLOR PALETTE / SKIN-TONE REFERENCE

CONSISTENCY:
Every panel must show the SAME person.

Keep facial identity, proportions, hairstyle, skin tone and physical characteristics consistent across every angle, pose and expression.

For views not clearly visible in the reference, intelligently reconstruct the character while staying strictly faithful to the visible reference. Do not invent major distinguishing features.

PRESENTATION:
Premium professional film-production character bible / modeling reference sheet.
Clean editorial layout.
Neutral studio background.
Soft realistic studio lighting.
Sharp facial details.
Natural skin pores and texture.
Individual hair strands.
Realistic fabric and accessories.
Accurate human anatomy.
Consistent camera perspective.
Clean thin borders between panels.
Professional labels underneath each view.
Minimal elegant typography.
High-end fashion/model reference aesthetic.

PHOTOREALISM:
Ultra-photorealistic.
Real human appearance.
Natural skin texture.
Realistic eyes.
Physically accurate hair.
Natural shadows.
Realistic proportions.
No CGI appearance.
No plastic skin.
No doll-like face.
No artificial beauty-face replacement.

NEGATIVE PROMPT:
different person, different face, face morphing, identity drift, altered facial structure, different eyes, different nose, different lips, different jawline, different skin tone, different hairstyle, inconsistent body proportions, duplicate person, distorted anatomy, asymmetrical face, plastic skin, waxy skin, CGI, 3D render, cartoon, anime, illustration, over-beautification, excessive makeup, unrealistic eyes, blurry face, low detail, extra fingers, missing fingers, malformed hands, bad anatomy.

OUTPUT:
One complete, highly detailed professional CHARACTER CONSISTENCY SHEET in a single image, suitable as a master reference for generating consistent images and videos of this character.

[Lucas @lksmlabc]

兄弟们，AI视频制作这赛道我是真有点跟不上了。

刚翻到一个叫OpenMontage的开源项目，GitHub已经冲到了当天热榜第一。它不是帮你剪，是压根不用你动手——丢一句人话进去，脚本、素材、配音、字幕、成片全给你跑完。

更狠的是，你丢一条自己喜欢的视频进去，它能拆掉别人的结构和节奏，重新给你搭一套制作方案。

看完第一反应：那我学剪辑图啥。做短视频的真得盯一下这玩意儿。
🔗 https://t.co/JduKMzUmPz

[Dan McAteer @daniel_mac8]

Astra's new compaction is experimental and is off by default. You must enable it.

Type this prompt into Codex.

Allows Astra to:

1. Save notes across context windows

2. Search messages and tool calls in earlier context windows

[Anshu @anshuc]

dude GPT-6 Astra is some kind of turbo-AGI machine god for 3D games. It one-shot this in 45 minutes for hardly a couple % of my quota.

I figured out how to get great graphics out of it. The trick is image gen. I'll share the process below.

[Nova @Nova0xz]

🔴 DESCARGAR CUALQUIER COSA DE INTERNET ACABA DE VOLVERSE ABSURDAMENTE FACIL

Alguien publicó una herramienta open source que descarga vídeos, audios, fotos y gifs de prácticamente cualquier web.

Solo pegas el enlace y ya está.

Sin anuncios. sin trackers. sin paywalls. sin historias.

No por nada tiene 42.3k estrellas en GitHub.

Se llama cobalt.

Te dejo el link en los comentarios 👇

[carpe @ftcarpe]

este tipo acaba de exponer cómo está usando Claude Code + YouTube para ganar $97k/mes.

menos de 15 minutos que valen la pena ver, guardalo para más tarde🙇🏻‍♂️

[Samuel Zeng @SamuelZengML]

Audio8 TTS Preview 0.1B — now on Hugging Face as ONNX INT8

~100M params · 11 languages · zero-shot voice cloning
Runs on CPU only (~0.4 GB RAM) — no PyTorch, no CUDA
44.1 kHz · OpenAI-compatible API · Apache 2.0

Try it 👇
https://t.co/lI7TFGZnjD

#TTS #VoiceCloning #ONNX #OpenSource

[Framer 🇱🇹 @Framer_X]

This is the best Seedance hack ever.

You can convert any 16:9 video to 9:16 with just one prompt.

Meaning you can repurpose all your horizontal videos as vertical IG Reels, YouTube Shorts, or TikToks 🤯

Prompt below 👇

[火山哥🕊️ @huoshan007]

兄弟们，AI剪视频这赛道已经卷到我看不懂了。

刚刷到一个开源项目，GitHub直接干到4.2万星。它不是帮你剪，是压根不用你动手。

你丢一句人话进去，脚本、素材、配音、字幕、成片，全给你跑完。

最离谱的是，你丢一条爆款视频进去，它能把人家的脚本结构、镜头节奏、画面风格全拆了，然后给你出好几套复刻方案。

底层12条流水线、100多个工具、700多个Agent技能包。这哪是工具，这是直接塞给你一个制作团队。

我昨天试了一下，第一反应是——那我学剪辑干嘛？说实话有点被吓到。

链接我甩这：https://t.co/gQmnzc3A1y，你们自己看。

反正我已经丢进收藏夹了，用不用再说。但做短视频的兄弟们，这玩意儿你们真得盯一下，不然明年可能真不用自己动手了。

[Wildminder @wildmindai]

Minimax H3 Latent Upscaler.
upscaler for 24-channel H3 video latents.

- fast low-res gen followed by high-res refinement
- skips VAE round-trips to save generation time
- prevents ghosting and double-image artifacts

https://t.co/Jk7znpcaz0

[angel @angeldot_]

GOOGLE HA LIBERADO EN SILENCIO UNA IA QUE PREDICE PATRONES

Ventas. Precios de mercado. Tráfico web.
Demanda energética. Volatilidad cripto.

Se llama TimesFM:

→ Entrenada con 100B de datos reales
→ Forecasting zero-shot, sin fine-tuning
→ Corre en local.

100% Gratis y Open Source.

Enlace abajo👇

[AI Evolution @AiEvolutio58513]

Stunned by a post: a 21-year-old American college student made $43,000 in a month with an AI virtual girlfriend.

He set up an OnlyFans account called Maya. This Maya is 22 years old, tagged as a UCF psychology dropout, with 1,247 paid subscribers, and the highest-spending fan shelled out $1,847. But the most crucial part? The girl in those photos doesn't even exist.

No real person on camera, no real person typing. All the messages are written by AI, the photos are AI-generated, and the voice is AI-synthesized. The entire Maya boils down to just four documents on his MacBook.

Back in the day, building an AI virtual influencer took anywhere from six months to a year and a half. Now, he's got it running in just 4 weeks. Looking ahead, it might take only a weekend to whip one up.

These four files break down like this:

· persona.md, spelling out Maya's character, backstory, and likes;
· voice.md, locking in her speaking tone and voice style;
· flux.md, standardizing her appearance and photo style to ensure all images look like the same person;
· brain.md, logging what each subscriber has chatted about and their preferences.

Before the AI replies to any fan, it reads through all four files, so the persona never breaks, and it never forgets what the fan has said.

Truth be told, the tech stack is so ridiculously simple it's almost absurd. That famous virtual influencer Aitana López from before? Her team took 18 months to pull her together. Now, a Maya-style setup is done and dusted in 4 weeks.

This isn't sci-fi—it's something anyone with a computer can replicate step by step. And it's not just OnlyFans; slap this workflow onto Instagram, TikTok, or Twitch, and it'll run just as smoothly.

AI has basically driven the cost of creating a virtual person down to zero. At this point, the only real barrier left is whether you've got the taste to decide what this virtual person should even look like.

REPOST AND FOLLOW @AiEvolutio58513 for more AI guides and tutorials like this.

[WaveSpeedAI @wavespeed_ai]

Seedance 2.5. Wan 3.0. MiniMax H3.
Three of the hottest video models right now.
Same test: emotion + acting.

Who delivers the most nuanced emotions?
Who gives the best performance?
You be the judge.

[roman @Nozelcode]

HAN RESUELTO EL MAYOR PROBLEMA DE CLAUDE CODE Y NADIE ESTÁ HABLANDO DE ELLO

Tu agente revisa tu proyecto entero cada vez que le escribes, y luego lo olvida.

Abre un archivo, ve que no era, abre otro.

En cada tarea, todo el rato.

4 de cada 10 tokens que pagas se van en eso.

Graft lo lee una sola vez y escribe lo que entendió: una ficha por cada parte del sistema, explicando qué hace y con qué se conecta.

Se quedan ahí, dentro de tu repo, como archivos normales.

Tu agente las lee antes de tocar nada. Llega sabiendo dónde vive cada cosa.

162 runs controlados:

→ 42% menos tokens
→ 46% menos tool calls
→ 60% menos tiempo
→ 32% menos coste
→ misma correctness

Sin embeddings. Sin vector DB. Corre en local.

OpenSource (MIT)
100% gratis

Enlace al repo abajo👇