Application Critique on Prisma

I am really interested in this application since I have ever encountered problems trying to convert pictures into a more artistic style using photoshop and Adobe Illustrator. Instead of drawing lines and covering colors, with one click Prisma is able to generate a very nice photo, which is quite amazing.

For the application, I think the following points are important from the presentation:

  1. Privacy issue
    1. Because this application makes use of A.I. and machine learning, for certain filters they are not able to handle it locally. When users upload the photo to the server and modify it, personal information is leaked. Therefore, here comes a privacy concern about how should Prisma protect users’ personal information and should Prisma use users privacy for commercial purpose.
  2. Community
    1. I could remember when group 9 was presenting 53, they introduced that 53 has a community that allow users to great artworks together. Group 5 also think that if they are going to implement this app, they are definitely to implement sharing features. In my opinion, sharing and having a community are the most important features that would keep users using this kind of applications. When people share creative artworks on social media or community, they would receive appreciation, which will encourage them keep using the app.
  3. Virtual Reality
    1. Virtual reality is getting popular and the market for it is growing. As for a photo editing application, Prisma actually have very strong power and should not just limit it in the current market. Possibly Prisma can improve its technology so that it is able to modify videos for virtual reality. Then people would be able to experience different stylistic sceneries and perhaps this would bring Prisma a lot of profits.

About my personal thoughts, I would like to criticize some thing first:

  1. I am not really sure whether it is reliable for the picture on slide 5, but currently in Singapore’s app store Prisma is not really the first in the “photo&video” category. It might not be very suitable to say “Prisma” is the first in industry.
  2. For its potential in movie industry, as far as I am concerned, it might not be really successful because for movies they have their own team of special effects and they are using more advanced tools to make sure the stylistic effects works well (as group 5 have mentioned that sometimes the picture is not generated well).

For my personal plan of implementing this application is:

  1. Make it a plug-in. People sometimes feel troublesome to open an application and just for editing photos (since this is not a social media and do not have very large user community). If it is possible, we can make Prisma a plug-in for Facebook, Instagram Twitter or WeChat. When people are having a post, they can directly use the plug-in inside social media to edit the photos. It would be more convenient for users and makes sure that Prisma is always being used.
  2. Create another version for those startup application designers or web-developers. The application would allow them to generate beautiful stylistic photos or icons. This would be very useful because sometimes people cannot find a good designers and using photoshop to edit every photo is really tedious. Prisma would help those people to save a lot of time.

 

Advertisements

8 thoughts on “Application Critique on Prisma

  1. Hi Yang Zhouhan,

    Glad you had some interesting takeaways from the presentation.

    “Prisma is not really the first in the “photo&video” category” Correct! But not completely. It is the first app to run neural networks to recreate photos. The photo you saw on slide 5, was from the app store from Russia hence the Russian language on that. In fact it was among the top 10 on the App Store in 77 countries few weeks back. Link: http://saudigazette.com.sa/technology/prisma-fever-grips-world/

    Plug-in idea brilliant, but usually Facebook or Twitter as we have seen don’t depend heavily on 3rd party apps (cues Twitter destroying Meerkat) so it might be a challegne unless it gets acquired by them.

    Regarding “Potential in movie industry” and “(as group 5 have mentioned that sometimes the picture is not generated well)”, I think there was a miscommunication. First, in terms of movie post-production, since I have decent experience in it, I know how effectively it can be used to post process the videos and to get that surreal feeling by just simply running that neural networks. It takes production houses like LucasFilms in Singapore, 3-6 months to edit a film (including both video and audio) Neural networks can dramatically bring down the cost and time required for it.
    For the picture quality not being good, we mentioned to a question asked in class why can’t we create our own filters. It is then we mentioned it takes time for the machine to learn and it needs training data to actually get used to a filter to apply it.

    Hope I made some sense. Happy to discuss more! (:

    Liked by 1 person

  2. Hi Zhuohan,

    I believe that implementing Prisma as a plugin would be really useful in making it a much more popular app. I’m not really sure how it will work though. It might not be easy to integrate its photo editing functionality to say Instagram.

    I believe the difficult thing for Prisma though is moving beyond a simple filter application app. Right now, Prisma’s main selling point is its ability to turn photos into art using Deep Learning. But this application really doesn’t solve any real pain point. Perhaps it answers the question of “how would Picasso paint what I am currently seeing” but I don’t foresee this question coming up too often.

    It is, however, interesting to think about Prisma’s technology. As discussed during the presentation Q&A, there are 2 main ways that the Deep Learning network could be implemented. One way would be a single input (i.e. the photo) and the output would be the transformed image. Another way would be to take two inputs (the photo and the filter) and the output would be the transformed image. The first way of doing would take much more time and is much more specific since every filter would have to have its own Deep Learning network. Of course, it is possible to speed up this process by ensuring that the first few layers of the Deep Learning network stay constant (since these layers usually extract lower level features) and only modify the back layers during training. The second way, which I believe is the correct way to solve the problem, would be much more general. While, in the first method, a Deep Learning network learns how to apply a specific filter, in the second method, a Deep Learning network learns instead to extract styles and content separately before merging them together.

    Another interesting question would be how Prisma comes up with their training data to train their neural network. If supervised learning were to be used, it would require a lot of painstaking effort in developing the correct outputs for a set of training data. I would really love to find out where the data comes from for their training phase.

    Liked by 1 person

    1. I suppose Prisma won’t use supervised learning, rather, reinforcement learning where the score is partially based on users’ action after seeing the result picture is more sound to me. 🙂

      Like

  3. Hi Bill,

    I think the plugin idea that you suggested isn’t very practical on mobile. I’m not sure if there are ways to make apps redirect to other apps. But yeah, its a good idea if its a web based plugin and it is well integrated into the workflow of sites like Twitter and Facebook.

    Personally, I think a photo is nice if it captures the people that you have a meaningful connection with. I don’t think I need a filter to remember the subjects in the photo better. After all it is a snapshot in time and you will never be able to retake that same image, with the same group of people, at that age in your live, etc. But surely Prisma serves a purpose.

    Nice suggestion on virtual reality being incorporated into Prisma. Whether it is technologically feasible is something that we have to wait and see, because it might simply be too draining on the resources of a mobile phone.

    CHEERS
    Derek

    Liked by 1 person

  4. Hi Zhuohan,
    I agree with you that the team stated “Prisma is the first in industry” but did not provide evidence to support that. There are a few similar app such as PicsArt, Pixlr.
    I do not quite agree with you about your plan of implementation. For me, Prisma itself is extremely cool and deserve to be a standalone app rather than some plugin to Facebook or Instagram. I believe people do not use the app just for the sake of posting to social media.
    For me, if there is something I want to improve in Prisma is making a web application for it. Right now Prisma is only available as a mobile app, which makes sense because people can take pictures on their phones. Adding a web app will be cool and reach out to more users who are not phone guru like myself XD.

    Liked by 1 person

  5. Hi Zhouhan,

    I do agree with you about the privacy when using Prisma. Despite the fact that my photo is processed by the machines and algorithm only, they are actually using my personal data. However, the point is that users do not really care about that until something happens.
    About your idea to make Prisma to become a plug-in, I think it is possible only if Facebook or Twitter buys Prisma, since they will not want to let a third-party app to be integrated in their main platform. Furthermore, I think that if the Prisma team goes for that idea, then it is not really a right move for them, since it does not add value to Prisma. They have their user base already, making Prisma to be a plug-in for Facebook or Twitter only brings more users to Facebook and Twiteer, not Prisma.

    Liked by 1 person

  6. There is a difference between the first photo maker application and the first photo maker application incorporating machine learning.
    I guess privacy is not such a major concern nowadays because most of the users who uploaded the images have a high tendency of sharing them with others (although this does’t justify the possible privacy leakage…); however, what I found disturbing is that all copy right of processed images belongs to the company.
    Community might be a good suggestion, and it is certainly important to support sharing function, where user can upload the processed image to other social media platforms such as Facebook, WeChat or Instagram. Without sharing function, user have to manually save the processed image and upload them to other social platforms again. Supporting this sharing feature not only make it easier for the user to share their joy with others, but can advertising Prisma further as well (through header or watermark).
    The other potential usage of current technology might be possible in the future, but I don’t see how Prisma would implement them now, and I guess they have more important features to support. 🙂

    Liked by 1 person

  7. Hi Yang Zhouhan,

    Thanks for your comment for our presentation.
    For our point of “first in industry”, we want to say that even though for the photo & category, Prisma is not the first one as you said, it is correct. However, for this kind of idea (use AI and ML to turn the ordinary photo to art) and the “supper power” it can give to you, we want to say that it is the one in the industry which can do that.
    For the plugin, making Prisma as a “plugin”, is veray good. Because the idea of Prisma is very simple, it can not engage user to use it for a long time, so it is better for Prisma to be the 3rd party of others daily app like Facebook and Twitter. However, it is not easy for Prisma because they need to negotiate with these comparies for the contract, not just easy as you said.
    For your point of movie industry, what we want to say is that we can combine a normal movie with the effect of Prisma to make it brilliant, instead of make in the one which is itself has its own effect. We dont want to let Prisma blurs the effect of the movie when adding artist effect to it, we want to make the normal movie (actually not a movie with too much content, we can consider as a short video) more beautiful with the effect of Prisma.
    For your last point, I am a bit agree with you, but I think that you are going too far away. Prisma now is very simple, and actually it is more for fun. Photoshop and Illustrator are more likely for “professional people”, and I think these are much more powerful than Prisma if Prisma want to do some kind of think like you said.

    Liked by 1 person

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s