3 ms·

Their example videos: https://openai.com/sora/ https://openai.com/sora/, of the doors opening, are hilarious. 1. The first set of doors doesn't have any doorkn

Their example videos: https://openai.com/sora/ https://openai.com/sora/, of the doors opening, are hilarious. 1. The first set of doors doesn't have any doorknobs or handles. https://ibb.co/PwqfzBq https://ibb.co/PwqfzBq 2. The second set of doors has handles, and some very large/random hinges on the left door. https://ibb.co/JkDtc6r https://ibb.co/JkDtc6r 3. The third set doesn't have any handles, but I can forgive that, because we're in a spaceship now. The problem is that the inside of the doors seem to have windows, but the outside of the doors, doesn't have any windows. https://ibb.co/nwpXmtq https://ibb.co/nwpXmtq & https://ibb.co/wr6v2g1 https://ibb.co/wr6v2g1 4. The best/most hilarious part for me. The doors have handles, but they are on the hinge side of the door. No idea how this would work. https://ibb.co/gWXDcfr https://ibb.co/gWXDcfr

ckcheng 2y ago
I think they trained on one too many closet bifold doors [1]. If you look at the edge of the doors as they swing open, it seems their movement resembles bifold door movement (there's a wiggle to it common to bifold doors that normal doors never have). Plus they seem to magically reveal an inner fold that wasn't there before. [1]: https://duckduckgo.com/?t=h_&q=interior+bifold+closet+doors&iax=images&ia=images https://duckduckgo.com/?t=h_&q=interior+bifold+closet+doors&...
neop1x 2y ago
There are more examples of its limitations. The video with dogs shows three taxis transforming into one, the number of people under the tree changing https://player.vimeo.com/video/1037090356?h=07432076b5&loop=1&controls=1 https://player.vimeo.com/video/1037090356?h=07432076b5&loop=... An example from the HunyuanVideo is terrible as well. Look at that awful tongue: https://hunyuanvideoai.com/part-1-3.mp4 https://hunyuanvideoai.com/part-1-3.mp4 And what we see in that marketing is probably the best they could generate. And I suppose it took a lot of prompt tweaking and regenerations. The internet is already full of junk shorts and useless videos and soon there will be even more junk content everywhere. :(

ckcheng 2y ago

I think they trained on one too many closet bifold doors [1]. If you look at the edge of the doors as they swing open, it seems their movement resembles bifold door movement (there's a wiggle to it common to bifold doors that normal doors never have). Plus they seem to magically reveal an inner fold that wasn't there before. [1]: https://duckduckgo.com/?t=h_&q=interior+bifold+closet+doors&iax=images&ia=images https://duckduckgo.com/?t=h_&q=interior+bifold+closet+doors&...

neop1x 2y ago

There are more examples of its limitations. The video with dogs shows three taxis transforming into one, the number of people under the tree changing https://player.vimeo.com/video/1037090356?h=07432076b5&loop=1&controls=1 https://player.vimeo.com/video/1037090356?h=07432076b5&loop=... An example from the HunyuanVideo is terrible as well. Look at that awful tongue: https://hunyuanvideoai.com/part-1-3.mp4 https://hunyuanvideoai.com/part-1-3.mp4 And what we see in that marketing is probably the best they could generate. And I suppose it took a lot of prompt tweaking and regenerations. The internet is already full of junk shorts and useless videos and soon there will be even more junk content everywhere. :(