Brothers, simply generating pictures can no longer satisfy me, so I started playing with minimax H3. There are several problems that bother me.
1. The open source version does not have 2k. Using three-party super-resolution tools, a 15-second video takes dozens of minutes. 2. Use qwen 3.8 27B jailbreak version for prompt words, and use qwen image 2.1 for pictures. There is no limit in itself, but minimax H3 will refuse.
Qwen Image 2.1
+ Gemini 3.8 Flash Lite TTS + LTX-2.5 experiment. The point is "First determine the voice, then create a video that matches the actual length of the voice." Therefore, all the audio is included in the video. IA2V (Image-and-Audio-to-Video) The first experiment was a success.
This has been asked of me
UbaldoHervas . The prompt is yours. All local AI Qwen 3.8 27B FP8. Use my own video engine, yes. But it's local AI... Prompt in comment
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. most “AI plays games” demos cheat with local algorithmic fallbacks
You have reached the end of the archive
All of qwen38