New AI matches GPT-5, new top image models, AI quests, video to 3D - AI NEWS
Recent advancements in AI include open-source models rivaling top closed models, Nvidia's scene lighting predictor, a new 4K image generator, Tencent's anime background remover, a robust speech-to-text AI, and Alibaba's FE2E tool excelling in depth and normal estimation with minimal training data.
MAIN POINTS FROM TRANSCRIPT
- New open-source models match top closed models like GPT5 and Gemini 2.5 Pro.
- Nvidia's AI predicts scene lighting for seamless object integration.
- Tencent Hunen's model excels in anime image background removal with 99.5% accuracy.
- Alibaba's FE2E tool accurately predicts image depth and surface orientation with minimal data.
TAKEAWAYS
- FE2E outperforms other models in depth and normal estimation with less training data.
- A new speech-to-text AI handles noisy audio and autodetects languages effectively.
- Stability releases a new music generator alongside other AI advancements.
- The latest image generator and editor supports 4K image creation for free.