v3: full multimodal pipeline — vision in, LLM recipe, TTS audio, FLUX.1 image generation 30f2a0a DevGruGold (Joe Lee) commited on May 10