Адобченко пилит свой Omnigen. И какой! Обещан код! В воскресенье мне негоже писать лонгриды

Адобченко пилит свой Omnigen. И какой! Обещан код! В воскресенье мне негоже писать лонгриды, поэтому я просто скопирую сюда все эти сладости, из которых меня больше всего возбудила вот эта: Learns from video data to understand shadows, reflections, lighting effects and object interactions Итак. * Universal Framework: Handle multiple tasks like generation, editing, customization and composition in a single model * Real-world Dynamics: Learns from video data to understand shadows, reflections, lighting effects and object interactions * Multi-Image Support: Process multiple input/output images seamlessly as “frames“ * Smart Context Understanding: Advanced hierarchical prompting system for precise control * Automatic Adaptation: Intelligently adjusts to different tasks without requiring specific instructions * Add/remove objects with natural lighting and shadows * Modify scene attributes while maintaining consistency * Seamless object insertion with proper perspective * Transfer objects between images with proper context * Adjust object poses and lighting automatically * Multi-subject composition with realistic interactions * Instruction Following: Better understanding of complex editing requests * Detail Preservation: Superior quality in maintaining important details * Natural Integration: More realistic blending of edited elements * Background Consistency: Better preservation of original image elements