Group discussed launching paid VIP membership with daily calls, announced contest winners, reviewed projects including AI animation and game development, and debated AI model complexity and language barriers. Captured meeting · August 1 · by Tu Michael (with help from AI, of course ^^ ) Key Takeaways - VIP community launching next week: $49/month with daily calls, recorded transcripts, and searchable database; all active members get free lifetime VIP access - Contest winners announced: Chris wins best Candy Crush clone and most useful app (Suno stem separator), Paul wins most creative, John wins best group member - Michael's AI animation placed top 50: Entered CapCut/Vietfilm AI Talent contest using Sora-generated clips stitched in CapCut, demonstrating professional-quality anime-style output - AI models are getting harder to prompt: Newer models use larger vocabularies with more ambiguous language, creating barriers for non-native English speakers and less experienced users Topics Discussed Speaker Diarization for 4,000 Videos Discussion about processing a large video archive to extract transcripts and tag speakers. Parallel processing on multi-core hardware could reduce the estimated 14-day serial timeline significantly. - Tools mentioned: Parakeet (IBM), Nemotron (Nvidia), Cole Medlin's $17/month service - Key challenge: Diarization (labeling who said what) is harder than transcription itself - Google Gemini advantage: Can tag speakers by login identity during live meetings, but backfilling archived video is a different problem. 1. Michael's AI Animation Work Michael demonstrated two anime-style short films created for a CapCut/Vietfilm AI Talent contest, placing 36th out of top 50 and advancing to the next round. Previously won top 3 in a smaller international contest when Sora launched. - Workflow: Generate 15-second clips in Sora with consistent characters using a "collage method" (face, full body, environment references), then stitch in CapCut