Introduction

A Gif (or Graphical Interchange Format) is a series of images or soundless videos that continuously loop without pressing the play button; such as short scenes from popular TV shows and movies. GIFs are a popular and fun way to communicate on the internet. GIFs can be better than a video because they usually have small file sizes so they don’t prevent web pages from loading quickly. They are also a great way to convey emotions in a fun way. Feeling excited to learn more? There’s a GIF for that.

Why this article?

I know there are few websites like giphy.com, Ezgif, gifs.com to turn a YouTube video into a GIF but these online GIF makers put watermark, have an upload limit of max file size 6MB, & don’t make GIFs longer than 10 seconds. Also, I want to explore on my own to find out how these websites are working under the hood.

In this article, we’ll walk you through 6 steps to make a GIF from a youtube video.

Table of Contents

1. Download youtube video using youtube-dl

2. Video post-processing/ crop video

3. Annotate video/ adding text using FFmpeg

4. mp4 to gif using FFmpeg

5. Add blur or grain filters to GIF

6. Optimize GIF using gifsicle

You can also refer to my YouTube video tutorial for better Understanding

Prerequisite

You require three libraries FFmpeg, Gifsicle, and Youtube-dl for this task. You can download the latest version from their official websites.

You can also usepip for installing written in python using the following commands:

pip install youtube_dl

pip install ffmpeg-python

pip install pygifsicle

If you want to follow along with me then please download command-line utilities (pre-installed executables files) from my GitHub Repository so that you don’t have to install anything.

First, open your command prompt and clone GitHub repo on your local machine

git clone [https://github.com/techykajal/youtube-to-gif-maker.git](https://github.com/techykajal/youtube-to-gif-maker.git)

Now navigate to the youtube-to-gif directory and unzip demo.rar file

cd downloads/youtube-to-gif

1. Download youtube video using youtube-dl

youtube-dl -o input.mkv [https://www.youtube.com/watch?v=AzjS8k5TEsY](https://www.youtube.com/watch?v=AzjS8k5TEsY)

• just grab the youtube video link to download the video using youtube-dl

• -o is an output key that follows the filename you want to save as

2. Video post-processing/ crop video

ffmpeg -i input.mkv -ss 30 -t 3 output.mp4

• Trim the video. This example will skip the first 30 seconds (-ss 30) of the input and create a 3-second output (-t 3).

Note- It is super amazing to see how quickly you can crop or trim your video without using any fancy heavy software like Premiere Pro or After Effects. To give you a brief comparison it took me 10 seconds to crop my video using a command prompt compared to 80 seconds in software. I know it’s not powerful like softwares but for simple edits, it is blazing fast.

3. Annotate video/ adding text using FFmpeg

ffmpeg -i output.mp4 -vf drawtext="fontfile=/path/to/font.ttf: \
text='KABOOM': fontcolor=white: fontsize=54: box=1: boxcolor=black@0.1: \
boxborderw=5: x=(w-text_w)/2: y=(h-text_h)/1.4" -codec:a copy output_text.mp4

• The @0.1 controls the opacity of the text box. In this example, it is set to 10%. You can remove @0.1 and there will be no transparency.

• -codec:a copy will stream copy the audio and avoid re-encoding.

• You can use ASS or SRT subtitles for timed text

• Please refer drawtext filter documentation for more options and examples.

4. MP4 to GIF using FFmpeg

ffmpeg -i output_text.mp4 -vf "fps=10,scale=480:-1:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" -loop 0 output.gif

• FFmpeg not only supports mp4, but you can also convert your video to GIF from other video formats like AVI, FLV, Mkv, WebM, & Ogg.

• change [fps](https://ffmpeg.org/ffmpeg-filters.html#fps) filter to set frame rate. In this example we are using 10 frame per second

• [scale](https://ffmpeg.org/ffmpeg-filters.html#scale) filter will resize the output to 480 pixels wide and automatically determine the height while preserving the aspect ratio. The Lanczos scaling algorithm is used in this example.

• [palettegen](https://ffmpeg.org/ffmpeg-filters.html#palettegen) and [paletteuse](https://ffmpeg.org/ffmpeg-filters.html#paletteuse) filters will generate and use a custom palette generated from your input.

• [split](https://ffmpeg.org/ffmpeg-filters.html#split) filter will allow everything to be done in one command and avoids having to create a temporary PNG file of the palette.

• Control looping with -loop output option. A value of 0 is infinite looping, -1 is no looping, and 1 will loop once meaning it will play twice. So a value of 10 will cause the GIF to play 11 times.

5. Add blur or grain filters to GIF

ffmpeg -i output.gif -vf fps=5,scale=480:-1,smartblur=ls=-0.5,crop=iw:ih-2:0:0 result.gif

You can skip this step if you don’t want to reduce the quality or size of the GIF

• -i is an input key that follows the filename you want to convert.

• -vf is key to set visual filters.

• fpsis a string describing output framerate (default “25”).

• scale sets the width and height, “-1” preserves aspect ratio.

• smartblur blurs or sharpens the video, “ls” means “luma_strength” — one of the smartblur options. Values from -1 to 0 sharpen the image.

• crop****trims pixels, “iw:ih-2:0:0” tells to trim 2 px from the bottom.

6. Optimize GIF using gifsicle

Optimize GIFs using the following command.

gifsicle -O3 result.gif -o result_optimized.gif

• O3 is an optimization level that tries several methods for better compression. Read Gifsicle documentation for more info.

• -ois an output key that follows with a filename.

Below is the final gif created using the above commands

Full Code

Link to GitHub Repository. Drop a star if you find it useful. ⭐️

#Command line script for youtube to gif maker
git clone https://github.com/techykajal/youtube-to-gif-maker.git

cd downloads/youtube-to-gif

youtube-dl -o input.mkv https://www.youtube.com/watch?v=AzjS8k5TEsY

ffmpeg -i input.mkv -ss 76 -t 4 output.mp4

ffmpeg -i output.mp4 -vf drawtext="fontfile=/path/to/font.ttf: \
text='KABOOM': fontcolor=white: fontsize=54: box=1: boxcolor=black@0.1: \
boxborderw=5: x=(w-text_w)/2: y=(h-text_h)/1.4" -codec:a copy output_text.mp4

ffmpeg -i output_text.mp4 -vf "fps=10,scale=480:-1:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" -loop 0 output.gif

ffmpeg -i output.gif -vf fps=5,scale=480:-1,smartblur=ls=-0.5,crop=iw:ih-2:0:0 result.gif

gifsicle -O3 result.gif -o result_optimized.gif

Conclusion

I’m thinking of using this set of code to make gifs from my youtube tutorials. I think having your own code makes your work super streamline and also gives lots of customization freedom than an online GIF maker.

What are you waiting for? Go make a GIF.

Interesting project ideas with source code and reference articles, also attaching some research papers too.

Are you excited to enter the Data Science world? Congrats! That’s still the right choice because of the ultimate boost in need of work done in Data Science and Artificial Intelligence during this pandemic.

Although, because of the crisis, the market currently gets tougher to be able to set it up again with more men force as they are doing earlier. So, It might possible that you have to prepare yourself mentally for long run hiring journey and many rejections in a way.

Hereby, while writing this article, I am assuming that you have already known that data-science portfolio is crucial and how to build it up. You might spend most of your time, doing data crunching and wrangling and not applying fancy models.

One question that I have asked on and on by data science enthusiasts is that what kind of projects should they include in their portfolio to build tremendously good and unique portfolio.

Below I have given the 8 unique ideas for your data science portfolio with attached reference articles from where you will get the insights of how to get started with any particular idea.

1. Sentiment analysis for depression based on social media post

This topic is so sensitive to be considered nowadays and in urgent need to do something about it. There are more than 264 million individuals worldwide who are suffering from depression. Depression is the main cause of disability worldwide and is a significant supporter of the overall global burden of disease and nearly 800,000 individuals consistently bite the dust because of suicide every year. Suicide is the second driving reason for death in 15–29-year-olds. Treatment for depression is often delayed, imprecise, and/or missed entirely.

Internet-based life gives the main edge chance to change early melancholy mediation services, especially in youthful grown-ups. Consistently, roughly 6,000 Tweets are tweeted on Twitter, which relates to more than 350,000 tweets sent for each moment, 500 million tweets for every day, and around 200 billion tweets for each year.

As indicated by the Pew Research Center, 72% of the public uses some sort of internet-based life. Datasets released from social networks are important to numerous fields, for example, human science and brain research. But the supports from a specialized point of view are a long way from enough, and explicit methodologies are desperately out of luck.

By analyzing linguistic markers in social media posts, it’s possible to create a deep learning model that can give an individual insight into his or her mental health far earlier than traditional approaches.

2. Sports match video to text summarization using neural network

So this project idea is basically based on getting precise summary out of Sports match videos. There are sports websites that tell about highlights of the match. Various models have been proposed for the task of extractive text summarization but neural networks do the best job. As a rule, Summarization alludes to introducing information in a brief structure, concentrating on parts that convey facts and information, while safeguarding the importance.

Automatically creating an outline of a game video gives rise to the challenge of distinguishing fascinating minutes, or highlights, of a game.

So, one can achieve that using some deep learning techniques like 3D-CNN (three-dimensional convolutional networks), RNN(Recurrent neural network), LSTM (Long short term memory networks) and also through Machine learning algorithms by dividing the video into different sections and then applying SVM(Support vector machines), NN(Neural Networks), k-means algorithm.

For better understanding, do refer to the attached articles in detail.

3. Handwritten equation solver using CNN

Among all the issues, handwritten mathematical expression recognition is one of the confounding issues in the region of computer vision research. You can train Handwritten equation solver by handwritten digits and mathematical symbols using Convolutional Neural Network (CNN) with some image processing techniques. Developing such a system requires training our machines with data, making it proficient to learn and make the required prediction.

Do refer to the below-attached articles for better understanding.

4. Business meeting summary generation using NLP

Ever got stuck in a situation, where everyone wants to see a summary not full reports. Well, I face it during my school and college days where we spend a lot of time preparing a whole report but the teacher only has time to read the summary.

Summarization has risen as an inexorably helpful way to tackle the issue of data over-burden. Extracting information from conversations can be of very good commercial and educational value. This can be done by feature capture of the statistical, linguistic, and sentimental aspects with the dialogue structure of the conversation.

Manually changing the report to a summed up form is too time taking, isn’t that so? But one can rely on Natural Language Processing (NLP) techniques to achieve that.

Text summarization using deep learning can understand the context of the entire text. Isn’t it a dream come true for all of us who need to come up with a quick summary of a document !!

Do refer to the below-attached articles for better understanding.

5. Facial recognition to detect mood and suggest songs accordingly

The human face is an important part of an individual’s body and it particularly plays a significant role in knowing a person’s state of mind. This eliminates the dreary and tedious task of manually isolating or grouping songs into various records and helps in generating an appropriate playlist based on an individual’s emotional features.

People tend to listen to music based on their mood and interests. One can create an application to suggest songs for users based on their mood by capturing facial expressions.

Computer vision is an interdisciplinary field that helps convey a high-level understanding of digital images or videos to computers. computer vision components can be used to determine the user’s emotion through facial expressions.

There are these APIs too that I found interesting and useful, although I didn’t work on these but attaching here with a hope that these will gonna help you.

6. Finding out habitable exo-planet from images captured by space vehicles like Kepler

In the most recent decade, over a million stars were monitored to identify transiting planets. Manual interpretation of potential exoplanet candidates is labor-intensive and subject to human mistake, the consequences of which are hard to evaluate. Convolutional neural networks are fit for identifying Earth-like exoplanets in noisy time-series data with more prominent precision than a least-squares strategy.

7. Image regeneration for old damaged reel picture

I know, how time- consuming and painful it is to get back your old damaged photo in the original form as it was earlier. So, this can be done using deep learning by finding all the image defects (fractures, scuffs, holes), and using Inpainting algorithms, one can easily discover the defects based on the pixel values around them to restore and colorize the old photos.

8. Music generation using deep learning

Music is an assortment of tones of various frequencies. So, the Automatic Music Generation is a process of composing a short piece of music with the least human mediation . Recently, Deep Learning engineering has become the cutting edge for programmed Music Generation.

FINAL WORD

I know that it’s a real struggle to build up a cool data science portfolio. But with such collection that I have provided above, you can make above-average progress in that field. The collection is new which gives opportunity for research purposes too. So, researchers in Data-Science can also choose these ideas to work on so that their research would be a great help for Data Scientists to start with project. And moreover It’s a real fun to explore the sides that nobody has done before. Although, this collection is actually constitute of ideas from beginning to advanced level.

So, I will not only recommend this for newbies in the data science area but also senior data scientists. It will open many new paths during your career, not only because of the projects but also through the newly gained network.

These ideas show you the broad range of possibilities and give you the ideas to think out of the box.

For me and my friends, the learning factors, adding value to the society and the unexplored knowledge is important and the fun in a way is essential. So, basically I enjoys doing such projects that give us a way to gain immense knowledge in a way and let us explore the unexplored dimensions. That is our main focus when dedicating time to such projects.

I hope you guys will find this article informative & useful for you. Do share your thoughts about these project ideas in the comment box & do let me know about other cool ideas if you have any✌️

You can reach me via the following :

1. Subscribe to my YouTube channel for video contents coming soon here
2. Follow me on Medium
3. Connect and reach me on LinkedIn