Test MiniMax H3 na scenie w stylu Simpsonów z Trumpem
I teraz od razu mowie, że musze odpocząć trochę od modeli wideo i będę przez najbliższy tydzień skupiał się na modelach do programowania Ale do…
Meanwhile, we're playing an FPS generated by Qwen 3.8 Flash Next in Studio, fully generated and playable IN BROWSER.
Not Fable, not Astra, but a model with only 6B active parameters (!) We're adding generic networking support, soon. Roblox is cooked.
兄弟们，单纯的生成图片已经满足不了我了，所以我开始玩minimax H3, 这里面好几个问题困扰着我
1、开源版本没有2k, 使用三方超分工具， 一个15秒的视频要几十分钟 2、提示词走qwen 3.8 27B越狱版，图片走qwen image 2.1本身也没有无限制，但是minimax H3会拒绝
Qwen Image 2.1
＋ Gemini 3.8 Flash Lite TTS ＋ LTX-2.5 の実験。 ポイントは、 「声を先に確定させ、その音声の実際の長さに合わせて動画を作る」 なので、音声が動画内に全てきっちり入る。 IA2V（Image-and-Audio-to-Video） の第一弾実験、成功です。
Esto me lo ha pedido
UbaldoHervas . EL prompt es suyo. Todo IA local Qwen 3.8 27B FP8. Usa mi motor propio de videos , eso si. Pero es IA local... Prompt en comentario
Can Qwen 3.8 27B running at 2000 tokens/sec pilot flappy bird in real time with ZERO hardcoded heuristics?
Part 3 of turning cerebras + Alibaba_Qwen 3.8 27B into an autonomous OS: closed loop flight control.
Y esto lo he hecho con Qwen 3.8 27B FP8 , mientras le ponía el colacao a mi hijo por voz desde la cocina.
Opensource diary Day 3
Made with Terresect, Qwen 3.8 27b on RTX 6000. Progress is sloooowwww.
Playing with qwen 3.8 27B in local only
My goal is to unlock commodity inference.
Not just better local ai, but better ai on cheap commodity hardware... We have made a lot of progress, a P100 GPU (~$100).
Qwen 3.8 27B MLX 4Bit
Qwen 3.8 + ltx 2.5
3.8 Flash did this based on a single screenshot.
The main work was done in 2 hours 23 minutes, followed by about 1 hour of editing. threejs And models like this can be run locally. TheAhmadOsman
Since everybody is spamming those stupid reels.
Qwen 3.8 27B, Kimi K3, Mimo V2.6 flash and astra Xhigh. obviously astra won. Mimi was surprising. kimi did no audio and...qwen is qwen
For fun I tried giving this assignment to GLM 5.3 Flash (Left) and Qwen 3.8 27b ThinkingCap (Right).
Both decided to use Python with PIL and numpy. Much, much simpler, but Qwen's pretty decent for a 27b model, just hope it wasn't destroyed by…
You can reportedly test 9 major AI models with 1 million tokens before paying—useful for comparing outputs or checking API compatibility.
The available models are listed as: • Claude Opus 5.5 • Claude Fable 5.1 • GPT-5.6 Sol • GPT-5.6 Luna • Grok 4.6 • Qwen 3.8 Max •
Alex created *FIRE* using Alibaba_Qwen 3.8 Flash in qoder_ai_ide - a *Contra*-style action game.
The protagonist is modeled after my son, Alex ❤️ It features 46 enemies, a boss, and a fiery cinematic intro.
It is ALL local generated.
Yet another proof of concept. Don't judge it by the graphics quality/animation.
This is what a 180b model dancing looks like, qwen 3.8 flash next fp8 across 2x dgx spark, and the red boxes are the part that surprise me.
10 to 12 w per gpu at idle 30 to 35 w per gpu while it generates, peaks around 55 w never past 60°c on either box 45 tok/s fresh,