H3 Max by fal

22 user comments (21 YouTube + 1 ProductHunt), nearly all at +0 upvotes, so there is no vote-weighted signal. Sentiment is overwhelmingly enthusiastic: 'ABSOLUTELY INSANE', 'What a fucking time to be alive', 'the most insane thing I've seen someone do with OpenAI', and 'a new kind of media, a bridge between movies and games'. The dominant theme is speed awe — 'models become faster in generating then we can speak or think'. The single ProductHunt comment carries the core spec: fal released a post-trained MiniMax H3 variant called Max, '50× faster than the original while maintaining quality', generating a 15-second video in ~9 seconds, which 'unlocks entirely new possibilities, like building an endlessly running live stream'. But the same users document concrete flaws: one physics error is called out twice ('girl phased through the desk lmao', 'the woman just walked right through the desk @4:08'); one user says 'the latency with prompting needs to be improved'; and another notes missing skin detail that upscalers can add only at the cost of making you 'look a little less like you and more like someone else'. Sentiment history matters: one user 'hated Minimax when they stopped giving free videos' 1.5–2 years ago and only forgave them after this release — pricing/access resentment lingers. Awareness is shallow: 'I have no idea what H3 is, but this looks cool'. Other signals: a question about 'usage limits on the pro 20x with astra' (about the GPT-6 stack in the demo, not H3 Max itself), a request to open-source the mac app, and off-topic engagement (asking when the demo track 'Compute Walk' hits Spotify).

No aggregate ratings were found for this product during the last harvest.