Season Rovang | Flickr

Flike

flike

Hello everyone! For several years now, I have been writing various "assistant" telegram bots for myself in Python that handle various small routine tasks for me - notifying me about something, checking service uptime, forwarding interesting content from telegram channels and chats, and so forth. This is convenient because the phone is always at hand, and being able to fix something on the server without even opening my laptop brings me particular pleasure. In general, I have accumulated a lot of different small project templates that I want to share with HackerNoon readers. I'll say right away that the examples may be niche in terms of their application "as is", but I will mark those places where, by changing a few lines of code to your own, you will be able to reuse most of the developments for your projects. I completed this specific project a few days ago, and it has already brought me a lot of benefits. I work at a Web3 infrastructure provider chainstack.com, dealing with a service for indexing data from smart contracts on EVM blockchains. And the quality of the service being developed critically depends on how "well" the nodes from which the service retrieves data online are functioning. I spent many hours trying to use ready-made tools that our infrastructure division uses, such as Grafana, BetterUptime, and others, but as I have little interest in the system's internals, with the main focus for me being the metrics at the entrance and the exit, I decided to write my own bot, which would do the following: At my request, it would go to the service, check the metrics, and send me a brief report on the current situation. At my other request, it would send me graphs of what has been happening over the last X hours. In case of a special situation, it would send me a notification that something is happening at that moment. In this article, I will focus on the first part, that is, receiving metrics on request. We will need a new virtual environment for work. cd ~ 
virtualenv -p python3.8 up_env  # crete a virtualenv
source ~/up_env/bin/activate  # activate the virtualenl Install dependencies: pip install python-telegram-bot
pip install "python-telegram-bot[job-queue]" --pre
pip install --upgrade python-telegram-bot==13.6.0  # the code was written before version 20, so here the version is explicitly specified

pip install numpy # needed for the median value function
pip install web3 # needed for requests to nodes (replace with what you need) File with functions (you can implement it with classes, but since the example is short, I did not plan to divide it into modules, but a multi-threading library requires functions to be moved to a separate file). Import dependencies: functions.py import numpy as np
import multiprocessing

from web3 import Web3 #  add those libraries needed for your task Describing a function for checking the state. In my case, it involved looping through pre-selected public nodes, retrieving their last block, taking the median value to filter out any deviations, and then, checking our own node against this median. Service state checking function (you can replace it with your own): # Helper function that checks a single node
def get_last_block_once(rpc):
    try:
        w3 = Web3(Web3.HTTPProvider(rpc))
        block_number = w3.eth.block_number
        if isinstance(block_number, int):
            return block_number
        else:
            return None
    except Exception as e:
        print(f'{rpc} - {repr(e)}')
        return None


# Main function to check the status of the service that will be called
def check_service():
    # pre-prepared list of reference nodes
    # for any network, it can be found on the website //chainlist.org/
    list_of_public_nodes = [
        '//polygon.llamarpc.com',
        '//polygon.rpc.blxrbdn.com',
        '//polygon.blockpi.network/v1/rpc/public',
        '//polygon-mainnet.public.blastapi.io',
        '//rpc-mainnet.matic.quiknode.pro',
        '//polygon-bor.publicnode.com',
        '//poly-rpc.gateway.pokt.network',
        '//rpc.ankr.com/polygon',
        '//polygon-rpc.com'
    ]
    
    # parallel processing of requests to all nodes
    with multiprocessing.Pool(processes=len(list_of_public_nodes)) as pool:
        results = pool.map(get_last_block_once, list_of_public_nodes)
        last_blocks = [b for b in results if b is not None and isinstance(b, int)]

    # define the maximum and median value of the current block
    med_val = int(np.median(last_blocks))
    max_val = int(np.max(last_blocks))
    # determine the number of nodes with the maximum and median value
    med_support = np.sum([1 for x in last_blocks if x == med_val])
    max_support = np.sum([1 for x in last_blocks if x == max_val])

    return max_val, max_support, med_val, med_support The next important file of the bot is . We import libraries and functions from the file above and set the necessary constants: uptime_bot.py import telegram
from telegram.ext import Updater, CommandHandler, Filters

from functions import get_last_block_once, check_service

# Here one can to set a limited circle of bot users, 
# listing the usernames of the users

ALLOWED_USERS = ['your_telegram_account', 'someone_else']
# The address of the node that I am monitoring (also a public node in this case)
OBJECT_OF_CHECKING = '//polygon-mainnet.chainstacklabs.com'
# Threshold for highlighting critical lag
THRESHOLD = 5 Next, let's describe a function that will be called when the command is issued from the bot's UI. def start(update, context):
    """Send a message when the command /start is issued."""
  
    try:
        # Get the user
        user = update.effective_user

        # Filter out bots
        if user.is_bot:
            return

        # Check if the user is allowed
        username = str(user.username)
        if username not in ALLOWED_USERS:
            return
    except Exception as e:
        print(f'{repr(e)}')
        return

    # Call the main function to check the network status
    max_val, max_support, med_val, med_support = check_service()
    # Call the function to check the status of the specified node
    last_block = get_last_block_once(OBJECT_OF_CHECKING)

    # Create the message to send to Telegram
    message = ""

    # Information about the state of the nodes in the public network (median, maximum, and number of nodes)
    message += f"Public median block number {med_val} (on {med_support}) RPCs\n"
    message += f"Public maximum block number +{max_val - med_val} (on {max_support}) PRCs\n"

     # Compare with the threshold
    if last_block is not None:
        out_text = str(last_block - med_val) if last_block - med_val < 0 else '+' + str(last_block - med_val)

        if abs(last_block - med_val) > THRESHOLD:
            message += f"The node block number shift ⚠️<b>{out_text}</b>⚠️"
        else:
            message += f"The node block number shift {out_text}"
    else: # Exception processing if a node has not responded
        message += f"The node has ⚠️<b>not responded</b>⚠️"

    # Send the message to the user
    context.bot.send_message(chat_id=user.id, text=message, parse_mode="HTML") Now, all that's left is to add the part where the bot is initialized, and the handler function is connected: token = "xxx"  # Bot token obtained from BotFather

# set up the bot
bot = telegram.Bot(token=token)
updater = Updater(token=token, use_context=True)
dispatcher = updater.dispatcher

# bind the handler function
dispatcher.add_handler(CommandHandler("start", start, filters=Filters.chat_type.private))

# run the bot
updater.start_polling() Finally, you can run the code on a cheap VPS server using: source ~/up_env/bin/activate
python uptime_bot.py After configuring the systemd unit file. As a result, the bot's work will look like this. If everything is fine: And if the lag becomes too large, then as follows: In the following articles, I will describe how to implement the two remaining tasks: Retrieve graphs on request showing the events that occurred over the last X hours. Receive an alert indicating that something is currently happening and requires action. The project's source code is available in the GitHub . If you found this tutorial helpful, feel free to give it a star on GitHub, I would appreciate it🙂 repository

How to Create a Telegram Bot to Monitor Your Service Uptime in Python (Part 1: Instant Metrics)

পাইথনে আপনার পরিষেবা আপটাইম নিরীক্ষণ করার জন্য কীভাবে একটি টেলিগ্রাম বট তৈরি করবেন (পর্ব 1: তাত্ক্ষণিক মেট্রিক্স)

So erstellen Sie einen Telegram-Bot zur Überwachung der Betriebszeit Ihres Dienstes in Python (Teil 1: Sofortige Metriken)

Cómo crear un bot de Telegram para monitorear el tiempo de actividad de su servicio en Python (Parte 1: Métricas instantáneas)

Comment créer un bot Telegram pour surveiller la disponibilité de votre service en Python (Partie 1 : Instant Metrics)

पायथन में अपनी सेवा के अपटाइम की निगरानी के लिए टेलीग्राम बॉट कैसे बनाएं (भाग 1: त्वरित मेट्रिक्स)

Python でサービス稼働時間を監視する Telegram ボットを作成する方法 (パート 1: インスタント メトリクス)

Python에서 서비스 가동 시간을 모니터링하기 위해 Telegram Bot을 만드는 방법(1부: 인스턴트 지표)

Como criar um bot de telegrama para monitorar o tempo de atividade do serviço em Python (parte 1: métricas instantâneas)

Python'da Hizmet Çalışma Sürenizi İzlemek için Telegram Botu Nasıl Oluşturulur (Bölüm 1: Anlık Ölçümler)

Cách tạo Bot Telegram để theo dõi thời gian hoạt động của dịch vụ của bạn bằng Python (Phần 1: Số liệu tức thì)

如何使用 Python 创建 Telegram 机器人来监控服务正常运行时间（第 1 部分：即时指标）

Audio Handling Basics: Process Audio Files In Command-Line or Python

Conceptos básicos de manejo de audio: Procesar archivos de audio en la línea de comandos o Python

In this tutorial, we’ll be creating a Tinder clone, in under 1 hour using , an open-source no-code platform that makes it seamlessly easy to create social applications. Phở Networks With Phở, you define your app in “graph” format (as shown below), and it generates the APIs in a breeze. Phở depends on , and ; so make sure you have them installed on your system before we start. PHP 7+ Redis Neo4J Redis Installation Instructions Neo4J Installation Instructions Phở uses Redis as its persistent (and lightning fast) keeper-of-truth, while Neo4J acts as the index with its query language. Thanks to this graph-native backend, you can query your Phở app in a language that’s in native graph format: Cypher MATCH (m)-[: ]->(f) m.sex="male" f.sex="female" f; like WHERE AND RETURN The query above would fetch all female users with at least 1 like. Needless to add, Phở is a declarative language. All you need to do is to define the core components of your apps (also called as, nodes and edges) in a whiteboard-friendly way, as follows: an Actor node User: a Subscribe edge Like: a Notification edge Message: Voila. So how do we implement this? Fortunately, there is a simple (graphical user interface) which lets you draw the application in a whiteboard friendly way as below: GUI To see how it works, take a look at the 15-minutes long tutorial video below: At the end, you’ll create a GraphQL recipe similar to: //github.com/esokullu/GroupRouletteRecipe We need to compile this and serve it via REST. To do that install pho-cli from . Github After you download the file, copy it to your global binary path such as on UNIX systems, so it can be called from anywhere. Also, make sure it’s executable by running: /usr/local/bin chmod +x local pho /usr/ /bin/ Then, while you’re on the terminal, run: > pho init This will start a CLI dialog with five easy questions: App Name App Description : You must select 6 (custom) here. It will ask you the path to your recipe. Either you can create it on your own as shown above, download, and enter that path; or just point to //github.com/esokullu/GroupRouletteRecipe for a quick start. App Template : keep it “admin” Founder Username Founder Password Founder Email Once you fill these out, it will create a folder with a bunch of files and folders, including the following key ones: this is the folder where we will place the GraphQL source code. schema/ this is the folder where compiled GraphQL schema will reside. build/ Then, we go to the root folder of our project, and install the dependencies using . Suppose, the App Name was “grouproulette”, here’s what we do: Composer > grouproulette cd > composer install We could have executed to compile our schema folder (in other words, the recipe) but it’s already baked into in this example, so we skip that. pho build Voila… Now we have a working-kernel, which we should pass to our REST API server, so we can connect it to our mobile app. If you need to change your server settings on the Phở side (your project folder) make sure to edit the .env file there. Please note it’s a hidden file so you may not be able to see it in your filesystem browser. Then run: > php public/api/index.php & Now if all goes well, when you visit on your browser, you should see a JSON response like: //localhost/founder { : , : , : } "success" true "id" "427501bd323e5e40f05b9fb4e769714a" "class" "PhoNetworksAutogenerated\\User" OK, now our REST APIs are working fine! For reference to the full set of commands, visit //www.phonetworks.org/assets/restapi.htm The default comes with the , , and functions in its . You may extend it further by editing the file as documented on Phở Server REST’s file. Phở Server REST login logout signup Authentication Controller public/api/index.php README To illustrate the signup call, a sample GET request would be: //localhost/signup?param1=joe&param2=&param3=secret&param4=01/15/1993&param5=&param6=//google.com/emre.gif&param7=0 Hope this helps! Should you have any questions, feel free to comment or shoot me an email at  Last but not least, if you need funding for the apps you create using Phở, we can help ! Previously published at //ideasforfree.org/2020/05/08/how-to-create-a-tinder-clone-in-pho/

How to Create a Tinder Clone Application in Phở [No-Code Approach]

As many of you may have been tracking, the Australian Parliament has been deliberating on a piece of legislation for the last couple of months called the . This bill seeks to give Australian law enforcement and intelligence agencies powers to force tech companies and telcos operating in Australia to do any number of things. Here’s a partial list: Telecommunications and Other Legislation Amendment (Assistance and Access) Bill 2018 (a) removing one or more forms of electronic protection that are or were applied by, or on behalf of, the provider; or (b) providing technical information; or © installing, maintaining, testing or using software or equipment; or (d) ensuring that information obtained in connection with the execution of a warrant or authorisation is given in a particular format; or (f) assisting with the testing, modification, development or maintenance of a technology or capability; or (h) modifying, or facilitating the modification of, any of the characteristics of a service provided by the designated communications provider; or (j) an act or thing done to conceal the fact that any thing has been done covertly in the performance of a function, or the exercise of a power, conferred by a law of the Commonwealth, a State or a Territory, so far as the function or power relates to: (i) enforcing the criminal law and laws imposing pecuniary penalties; or (ii) assisting the enforcement of the criminal laws in force in a foreign country; or (iii) the interests of Australia’s national security, the interests of Australia’s foreign relations or the interests of Australia’s national economic well-being. Pretty scary stuff, huh? We thought so too. We’ve been tracking the progress of the bill and have been trying to analyse it to work out what the implications might be for software companies both inside and outside of Australia, and especially for Loki. The Assistance and Access Bill 2018 (Shortened to AAA18 for the rest of the article) gives Australian agencies the ability to issue 3 types of notices to ‘communications service providers.’ The definition of ‘provider’ in the legislation is very broad. Pretty much anyone that provides any service or product that involves the internet could fall under its scope, and the notices that can be issued increase in scope and obligation, and are called Technical Assistance Requests (TARs), Technical Assistance Notices (TANs), and Technical Capability Notices (TCNs). The latter is a legally enforced instruction to create or modify features to give an agency a new technical capability. Although this notice must come from the Attorney General of Australia, the scope for new espionage tools to be created by this notice is extremely broad. The scariest thing about this bill is the penalties given to providers who leak information about the investigation or notice, or refuse to comply with the notice. Jail sentences as long as 10 years could apply to a whistleblower, and given that these notices could be issued to companies and individuals that provide services to Australians, these notices could be issued to almost anyone around the world. With strong extradition treaties, this legislation could reach people across the nations of the Five Eyes Alliance (UK, US, AU, NZ, CA) and beyond. Some, myself included, strongly suspect that this is a coordinated effort by the Five Eyes alliance to gain access to the world’s most popular applications, as the UK recently to the already controversial Investigatory Powers Act of 2016, so that it is now quite closely aligned with this new Australian legislation. They even use similar terms, with the main similarity being the Technical Capability Notice. pushed through an amendment Thankfully, AAA18 does explicitly state that these notices cannot be used to force a company to break its own encryption, introduce security flaws, or deliberately ignore existing flaws. It also explicitly says that these notices cannot be used to introduce a ‘systemic weakness’ into the product or service. There were a great many concerns about the definition of ‘systemic weakness’ being too vague in the legislation. The final amendment to the bill gave us the following definitions: systemic vulnerability means a vulnerability that affects a whole class of technology, but does not include a vulnerability that is selectively introduced to one or more target technologies that are connected with a particular person. For this purpose, it is immaterial whether the person can be identified. systemic weakness means a weakness that affects a whole class of technology, but does not include a weakness that is selectively introduced to one or more target technologies that are connected with a particular person. For this purpose, it is immaterial whether the person can be identified. target technology : (a) for the purposes of this Part, a particular carriage service, so far as the service is used, or is likely to be used, (whether directly or indirectly) by a particular person, is a target technology that is connected with that person; and (b) for the purposes of this Part, a particular electronic service, so far as the service is used, or is likely to be used, (whether directly or indirectly) by a particular person, is a target technology that is connected with that person; and © for the purposes of this Part, particular software installed, or to be installed, on: (i) a particular computer; or (ii) a particular item of equipment; used, or likely to be used, (whether directly or indirectly) by a particular person is a target technology that is connected with that person; and (d) for the purposes of this Part, a particular update of software that has been installed on: (i) a particular computer; or (ii) a particular item of equipment; used, or likely to be used, (whether directly or indirectly) by a particular person is a target technology that is connected with that person; and (e) for the purposes of this Part, a particular item of customer equipment used, or likely to be used, (whether directly or indirectly) by a particular person is a target technology that is connected with that person; and (f) for the purposes of this Part, a particular data processing device used, or likely to be used, (whether directly or indirectly) by a particular person is a target technology that is connected with that person. For the purposes of paragraphs (a), (b), ©, (d), (e) and (f), it is immaterial whether the person can be identified. What this effectively means is that providers can not be forced to make changes to their products that negatively affect every user of that product. Instead, they can only be ordered to create the means by which they could selectively inject weaknesses or vulnerabilities into a specific product in use by a specific targeted person. However, AAA18 gives the legislative authority for agencies to create and install monitoring tools and other intrusive mechanisms into all kinds of software and hardware. For this iteration of the bill, these tools can only be switched on and used to target crimes with minimum sentences of 3 years or more. However, there is nothing to say that won’t change and very little oversight is required by this bill. It is also extremely problematic that these tools will even exist in the first place. If they fall into the wrong hands, the effects will be devastating. The NSA in the US developed a range of surveillance techniques that were eventually leaked and used against American interests by criminals and foreign governments, and the same will likely happen here. The introduction of AAA18 and its UK equivalent means that applications such as WhatsApp, Signal, Facebook Messenger, Gmail, and any other popular communications medium can silently turn into a monitoring device for ASIO, ASDS, AFP, GCHQ, MI5, and so on. The companies behind these products can’t utter a peep about these notices or their repercussions, or warn their users. They are even given protections to indemnify themselves against any civil cases caused by any later discovery of eroded privacy. This bill does not require these companies to break the encryption of their systems, but there are plenty of other things these agencies could force companies to create and install, such as tools that would allow them to remotely pull information from a specific user’s device post-encryption, or gain access to these services’ servers where treasure troves of metadata could be harvested. Every ‘private’ messaging application out there could now be forced to send data back to intelligence agencies about who is communicating to who and when the communications are taking place. , and companies across the world can be compelled to follow them. This isn’t hypothetical anymore — these powers are now law We will not know how widely these powers will be deployed until the first annual report is released. All it takes is one TCN for each of Facebook, Google, and WhatsApp, and the vast majority of private communications will be corruptible at the whim of the UK/Australian Intelligence and Law Enforcement community. The information that they can gather from this can easily be shared across the Five Eyes Alliance, effectively giving these tools to the governments of the whole English speaking world. For advocates of digital privacy such as myself, this is a very concerning development. If anything, it only strengthens the need to shift the paradigm in communication tools so that they are decentralised, open source and private by default. If we succeed, laws such as these can’t dissolve our ability to collectively access private spaces online. I have said this before, but access to these online spaces is critical to a healthy 21st century democracy, and our governments are too quick to dismiss the need for widespread digital security. How it will affect Loki Obviously, we were terrified when we first saw this bill. The potential for the project to be entirely undermined by this legislation did not go unnoticed. We had begun to consider how we might set up failsafes to allow people to catch bad code being injected into our codebase, or to pay someone external to Loki to do regular inspections of our binaries that we release and ensure they are not leaking extra information or mismatching the codebase in some way. If we were to be issued a TCN, we would not be able to tell anyone about it. If we set up some sort of canary system, we could be imprisoned. So whatever failsafe we did set up would have to be external to Loki, and would have to be regularly auditing us to make sure we haven’t been compromised before a TCN was issued. We had also considered that we may have to leave Australia altogether. However, given the legislation allows them to target any company whether they are in Australia or not, and given most of the Loki team currently lives in Australia and has family here, this would be a very extreme and difficult move to make. None of us would ever be able to return home if we were issued a notice and refused to comply. We may have simply been extradited out of wherever we moved to anyway, so this option seemed extreme and ineffective. New amendments to the bill make this move unnecessary. Our analysis of the bill and its proposed amendments lead us to the following conclusions: With the addition of the amendment defining ‘systemic weakness,’ any modification to the Loki source code that gives authorities new capabilities would almost certainly be classed as a systemic weakness. Thus, we see it is a nearly impossible that we would be forced to implement any privacy degrading code into the software that we release to the public. It is feasible that we may be forced to develop an alternative client software for authorities to use with additional data collecting features or something of the sort. Thankfully, given the network is regulated by Service Nodes, the authorities would still need to own 40%+ of the entire Service Node network to make this effective for widespread surveillance. Given our economic assumptions as outlined in our , this is likely to be prohibitively expensive for the government, particularly if the usage of Loki is so high that it is worth the government issuing a TCN in the first place. Cryptoeconomics proposal As long as we are able to keep our code open source, we can guarantee that our code will always be auditable. As such, it may make sense to go over our existing licenses and convert them to GPLv3 to prevent closed source software derived from our implementation from appearing. We will require written permission from the copyright holders of the projects we have forked in order to achieve this. As Loki is a decentralised system with extensive privacy protections, there is very little we as the Loki team can do to de-anonymise our users even if we wanted to. The extent of the information we might be able to provide to authorities are all known attacks such as DPI and traffic shape correlation, however we don’t have any additional information at hand than any other node operator on the network. This means that if we are issued a notice, we won’t be much use to these agencies anyway. The chances that Loki will eventually be issued a notice are fairly high, however such a notice would not result in compromising the privacy that our system provides. The same can not be said of other communication platforms, where the service provider has central control over the data being transported. Your phone number is attached to your WhatsApp and Signal accounts, and all of the metadata that you create is now up for grabs. Only by decentralising the routing and storage of communications can it become truly private under this new legislation. The case for a system like Loki has never been stronger. What you can do to stay private on the internet With this new legislation, it is important to understand and be vigilant about our privacy on the internet. Wherever practical, using open source software is a good start to protecting one’s privacy. Open source software is much more trustworthy than closed source applications as the code is generally reviewed by people form all over the world. If any backdoors or questionable features are added to this software, alarm bells should be ringing shortly afterwards. Whenever possible, you should also always build the open source software you are using from source. This will remove another layer of trust, as you’ll know the application you build is running exactly what is in the codebase you want to use. Nothing extra, nothing less. It is entirely feasible that modified versions of applications can put in place instead of the real ones if a notice compels a developer to do so. Reproducible builds help to mitigate against this, and this is something we are now striving towards, but it is your responsibility to check you didn’t get served a fake version of the application you want to use. VPNs are a decent first step in protecting network privacy, but this legislation could severely undermine their utility. If the VPN you use is compelled to install monitoring systems in their service, you might just be handing your browsing and connection information straight over to the authorities and be none the wiser. A better approach is to use some sort of mixnet. Tor is the obvious choice for the time being, but when Lokinet is launched, you should use it, as you’ll will be able to use essentially any program you like straight out of the box (it’s both TCP and UDP friendly). You’ll enjoy low latency, and a dedicated network of incentivised nodes that you can reasonably trust to not be largely comprised of nodes run by surveillance agencies. I hope this article has clarified the situation for you, and given you some insight to why decentralised privacy is so important. If you’d like to follow what we’re doing at Loki, head to //loki.network Originally published at loki.network .

Loki’s Response to the Assistance and Access Bill 2018

Emotion As A Service: AI’s Next Market Is Your Heart

العاطفة كخدمة: السوق القادمة للذكاء الاصطناعي هي قلبك

Емоцията като услуга: Следващият пазар на AI е вашето сърце

পরিষেবা হিসাবে আবেগ: AI এর পরবর্তী বাজার আপনার হৃদয়

La emoción como servicio: el próximo mercado de la IA es tu corazón

Emosyon kòm yon sèvis: pwochen mache AI a se kè ou

感情をサービスとして: AIの次の市場はあなたの心

អារម្មណ៍ជាសេវាកម្ម៖ ទីផ្សារបន្ទាប់របស់ AI គឺជាបេះដូងរបស់អ្នក។

Кызмат катары эмоция: AIнын кийинки рыногу - бул сиздин жүрөгүңүз

Emocje jako usługa: następnym rynkiem dla sztucznej inteligencji jest Twoje serce

Emotion seSevhisi: AI Inotevera Musika Mwoyo Wako

ስምዒት ከም ኣገልግሎት፡ ቀጻሊ ዕዳጋ AI ልብኻ’ዩ።

Cảm xúc như một dịch vụ: Thị trường tiếp theo của AI là trái tim của bạn

Tired? Burned Out? You Might Just Be Bored

Yorğun? Yandı? Sadəcə Darıxmış Ola bilərsən

Κουρασμένος; Κάηκε; Μπορεί απλά να βαριέσαι

¿Cansado? ¿Agotado? Puede que simplemente estés aburrido

Canso? Queimado? Podes estar aburrido

疲れている？燃え尽きている？退屈しているだけかも

Pavargote? Perdegė? Jums gali būti tiesiog nuobodu

Уморен? Изгорени? Можеби ќе ви биде досадно

Obosit? Ars? S-ar putea să te plictisești

Umechoka? Umechomwa? Unaweza Kuchoshwa Tu

சோர்வாக? எரிந்துவிட்டதா? நீங்கள் சலிப்படையலாம்

Charchadimi? Yondimi? Siz shunchaki zerikishingiz mumkin

Ndidiniwe? Utshisiwe? Unokuba Ukruqukile

Understanding Bloom Filters: An Efficient Probabilistic Data Structure

Growing up, my weekends weren’t spent at the mall or hanging out with friends like most high schoolers my age. I was trapped in my parents' store, surrounded by boxes, shelves, and countless products awaiting inventory. My job? Take photos of products, upload them to the website, and write descriptions that somehow convinced people to buy yet another brand of phone case or random kitchen gadget. It was tedious work—the same routine over and over. I felt like a robot, except the real robots weren’t around to help. Between balancing schoolwork and the constant cycle of restocking and updating the online store, I was burnt out. The worst part wasn’t even restocking itself; it was the endless product uploads and descriptions. By Saturday afternoon, after hours of taking and editing product images, I couldn’t tell the difference between a toaster and a blender. Describing them? Even worse. I finally decided there had to be a better way. I remembered reading about AI tools that could do things like generate text and images. That’s when I found Coze.com—a no-code platform that allows anyone to create an AI chatbot using natural language to prompt its system. It was like a light bulb flickered in my head. What if I could build a chatbot to handle all the boring stuff? This could help me save time I spent taking product images, manually researching, and tying in product details into our website. Well, this I did and will show you how. Below is an outline of what I’ll discuss on how I implemented its tools in building my AI chatbot What is Coze?
How it works
Writing Prompts
Implementing Plugins
Variables
Database
Outcome
Benefits of using my AI chatbot
Personal thoughts about Coze What’s Coze? In short, Coze is a no-code tool platform for building simple to complex AI chatbots in minutes. It allows users to easily create a wide range of bots using text. The chatbots can be published in its in-app store, WhatsApp, Telegram, Discord, Slack, and a whole other bunch of social media. How It Works To create your own chatbot, simply Sign up on the platform.
Create an agent bot in your workspace.
Fill in all the required information on the pop-up screen.
Begin prompting. Prompts When customers ask or search for an item on our website, they expect to see a detailed description of it. They want to have information on what they are buying, its quality, specifications, and what makes it stand out. To provide this, I have to instruct the system on the input data it should get, and output information it should return, and in what structure. To do this, I utilize the prompt feature. A prompt is a natural language instruction that tells an AI system what to do, and Coze uses this to make it easy for anyone to describe what they intend to build. These instructions are entered into the system under personas and prompts. In my case, I wanted my bot to take the name of a product and provide me with a detailed description and specifications of it. I noticed there are two types of prompting while building. Agent prompt, which tells the chatbot what to build, how to build it, and results to return to users. End user prompt, which queries the built chatbot to return desired results Coze allows for the optimization of initially written prompts so it can return the best results in a well-structured manner, including a well-detailed skill set my chatbot should have. Plugins As the saying goes, pictures speak louder than words, and what better way to reinforce the item specifications than giving customers an image of the item they searched for and intend to buy? They don’t just get to read about what the products are made of or can do but get a visual representation of every detail. To add this level of detail to my chatbot, plugins came into play. Plugins are what make most Coze bots run faster and more effectively, delivering the best results, much like how my parent’s store utilizes the services of established delivery companies to send out paid items to customers. They are additional software components that extend the functionality of an application. Plugins provide specialized features and capabilities that the base application, in this case, my AI chatbot, does not have. To enhance my chatbot, I made use of two plugins: Stable Diffusion and Gemini AI. Gemini AI provided my chatbot with information on inputted product items. It takes the name of the product from the user prompt, searches through its large knowledge base, and returns a response. A stable diffusion plugin makes it easy to generate high-quality product images. Database Much like how we have a warehouse for storage and retrieval of items. I needed to be able to store the product information generated for easy reference and retrieval instead of having to prompt the system every time I needed the same information. Databases are used to store information to be retrieved at a later time. For my chatbot, I created a database by specifying the table name, name of fields, type of data to be collected, and if it required information. Since Coze is a no-code tool, I can easily type in this field name, and it would be created as an SQL database. To record data in the database, I simply specified a new skill in the prompt and persona section, referencing the database table name and its fields. Also, I enabled a file box that, in Coze, holds large files, such as images, to store the product images. This would come in handy when I needed to download the images to use on my parent’s online store. Check out Skill 3 in the image below. Outcome In the preview panel, I tested my newly created chatbot and got the following results as shown in the image below. My chatbot generated a response when queried with a product name and image and stored the result in its database for easy retrieval when needed or the chat has been cleared. Benefits of Using My AI Chatbot By the time I had it fully functional, everything changed. Instead of spending hours with a camera taking product pictures and trying to give a description of the product based on my knowledge, I’d simply input the product name, and my AI chatbot would take care of the rest. It even handled multiple product variations. The best part is it helped cut down our marketing and advertising costs because the AI-generated images were good enough to post right away on social media and campaign boards. We didn’t have to spend huge amounts hiring a professional photographer to take product images for the store when we needed to run a sales campaign. Now, what used to take me an entire weekend only took a couple of hours. Personal Thoughts After building my chatbot utilizing Coze, here are my personal thoughts about it. Coze makes complicated AI chatbots look easy. Before them, you had to be an experienced developer to build out an AI product, but with Coze, a novice can get started easily with zero level of expertise. Making AI chatbot development more accessible. Launching your chatbot on its platform or another platform, such as WhatsApp, is simple with guides to help individuals achieve this. This makes using my developed chatbot on my favorite platform without hassle. Sometimes, it feels the plugins are limited but Coze allows for the development of plugins to be added to its library. Anyone can design a plugin and publish it on the Coze store. This alone allows for a more robust platform and tools. Coze's emphasis on using real-time data and personalization enables the creation of more tailored and effective AI chatbots. This improves user experiences and drives engagement with businesses and services. Coze encourages experimentation and innovation in AI chatbot development. This leads to the creation of new and exciting chatbot applications which addresses a variety of needs. Growing up, my weekends weren’t spent at the mall or hanging out with friends like most high schoolers my age. I was trapped in my parents' store, surrounded by boxes, shelves, and countless products awaiting inventory. My job? Take photos of products, upload them to the website, and write descriptions that somehow convinced people to buy yet another brand of phone case or random kitchen gadget. It was tedious work—the same routine over and over. I felt like a robot, except the real robots weren’t around to help. Between balancing schoolwork and the constant cycle of restocking and updating the online store, I was burnt out. The worst part wasn’t even restocking itself; it was the endless product uploads and descriptions. By Saturday afternoon, after hours of taking and editing product images, I couldn’t tell the difference between a toaster and a blender. Describing them? Even worse. I finally decided there had to be a better way. I remembered reading about AI tools that could do things like generate text and images. That’s when I found Coze.com—a no-code platform that allows anyone to create an AI chatbot using natural language to prompt its system. It was like a light bulb flickered in my head. What if I could build a chatbot to handle all the boring stuff? This could help me save time I spent taking product images, manually researching, and tying in product details into our website. Well, this I did and will show you how. Below is an outline of what I’ll discuss on how I implemented its tools in building my AI chatbot What is Coze? How it works Writing Prompts Implementing Plugins Variables Database Outcome Benefits of using my AI chatbot Personal thoughts about Coze What is Coze? What is Coze? How it works How it works Writing Prompts Writing Prompts Implementing Plugins Implementing Plugins Variables Variables Database Database Outcome Outcome Benefits of using my AI chatbot Benefits of using my AI chatbot Personal thoughts about Coze Personal thoughts about Coze What’s Coze? In short, Coze is a no-code tool platform for building simple to complex AI chatbots in minutes. It allows users to easily create a wide range of bots using text. The chatbots can be published in its in-app store, WhatsApp, Telegram, Discord, Slack, and a whole other bunch of social media. How It Works To create your own chatbot, simply Sign up on the platform. Create an agent bot in your workspace. Fill in all the required information on the pop-up screen. Begin prompting. Sign up on the platform. Create an agent bot in your workspace. Fill in all the required information on the pop-up screen. Begin prompting. Prompts Prompts When customers ask or search for an item on our website, they expect to see a detailed description of it. They want to have information on what they are buying, its quality, specifications, and what makes it stand out. To provide this, I have to instruct the system on the input data it should get, and output information it should return, and in what structure. To do this, I utilize the prompt feature. A prompt is a natural language instruction that tells an AI system what to do, and Coze uses this to make it easy for anyone to describe what they intend to build. These instructions are entered into the system under personas and prompts. In my case, I wanted my bot to take the name of a product and provide me with a detailed description and specifications of it. I noticed there are two types of prompting while building. Agent prompt, which tells the chatbot what to build, how to build it, and results to return to users. End user prompt, which queries the built chatbot to return desired results Coze allows for the optimization of initially written prompts so it can return the best results in a well-structured manner, including a well-detailed skill set my chatbot should have. Plugins Plugins As the saying goes, pictures speak louder than words, and what better way to reinforce the item specifications than giving customers an image of the item they searched for and intend to buy? They don’t just get to read about what the products are made of or can do but get a visual representation of every detail. To add this level of detail to my chatbot, plugins came into play. Plugins are what make most Coze bots run faster and more effectively, delivering the best results, much like how my parent’s store utilizes the services of established delivery companies to send out paid items to customers. They are additional software components that extend the functionality of an application. Plugins provide specialized features and capabilities that the base application, in this case, my AI chatbot, does not have. To enhance my chatbot, I made use of two plugins: Stable Diffusion and Gemini AI. my AI chatbot, Stable Diffusion Gemini AI. Gemini AI provided my chatbot with information on inputted product items. It takes the name of the product from the user prompt, searches through its large knowledge base, and returns a response. A stable diffusion plugin makes it easy to generate high-quality product images. Database Much like how we have a warehouse for storage and retrieval of items. I needed to be able to store the product information generated for easy reference and retrieval instead of having to prompt the system every time I needed the same information. Databases are used to store information to be retrieved at a later time. For my chatbot, I created a database by specifying the table name, name of fields, type of data to be collected, and if it required information. Since Coze is a no-code tool, I can easily type in this field name, and it would be created as an SQL database. To record data in the database, I simply specified a new skill in the prompt and persona section, referencing the database table name and its fields. Also, I enabled a file box that, in Coze, holds large files, such as images, to store the product images. This would come in handy when I needed to download the images to use on my parent’s online store. Check out Skill 3 in the image below. Outcome In the preview panel, I tested my newly created chatbot and got the following results as shown in the image below. My chatbot generated a response when queried with a product name and image and stored the result in its database for easy retrieval when needed or the chat has been cleared. Benefits of Using My AI Chatbot Benefits of Using My AI Chatbot By the time I had it fully functional, everything changed. Instead of spending hours with a camera taking product pictures and trying to give a description of the product based on my knowledge, I’d simply input the product name, and my AI chatbot would take care of the rest. It even handled multiple product variations. The best part is it helped cut down our marketing and advertising costs because the AI-generated images were good enough to post right away on social media and campaign boards. We didn’t have to spend huge amounts hiring a professional photographer to take product images for the store when we needed to run a sales campaign. Now, what used to take me an entire weekend only took a couple of hours. Personal Thoughts After building my chatbot utilizing Coze, here are my personal thoughts about it. Coze makes complicated AI chatbots look easy. Before them, you had to be an experienced developer to build out an AI product, but with Coze, a novice can get started easily with zero level of expertise. Making AI chatbot development more accessible. Coze makes complicated AI chatbots look easy. Before them, you had to be an experienced developer to build out an AI product, but with Coze, a novice can get started easily with zero level of expertise. Making AI chatbot development more accessible. Launching your chatbot on its platform or another platform, such as WhatsApp, is simple with guides to help individuals achieve this. This makes using my developed chatbot on my favorite platform without hassle. Launching your chatbot on its platform or another platform, such as WhatsApp, is simple with guides to help individuals achieve this. This makes using my developed chatbot on my favorite platform without hassle. Sometimes, it feels the plugins are limited but Coze allows for the development of plugins to be added to its library. Anyone can design a plugin and publish it on the Coze store. This alone allows for a more robust platform and tools. Sometimes, it feels the plugins are limited but Coze allows for the development of plugins to be added to its library. Anyone can design a plugin and publish it on the Coze store. This alone allows for a more robust platform and tools. Coze's emphasis on using real-time data and personalization enables the creation of more tailored and effective AI chatbots. This improves user experiences and drives engagement with businesses and services. Coze's emphasis on using real-time data and personalization enables the creation of more tailored and effective AI chatbots. This improves user experiences and drives engagement with businesses and services. Coze encourages experimentation and innovation in AI chatbot development. This leads to the creation of new and exciting chatbot applications which addresses a variety of needs. Coze encourages experimentation and innovation in AI chatbot development. This leads to the creation of new and exciting chatbot applications which addresses a variety of needs.

How to Build a Product Identifier AI Chatbot

Як стварыць чат-бот AI ідэнтыфікатара прадукту

Cómo crear un chatbot de inteligencia artificial que identifique productos

כיצד לבנות מוצר מזהה AI Chatbot

製品識別子 AI チャットボットの構築方法

របៀបបង្កើតផលិតផលកំណត់អត្តសញ្ញាណ AI Chatbot

제품 식별자 AI 챗봇을 만드는 방법

Tsela ya go Aga Sešupo sa Setšweletšwa AI Chatbot

Jak zbudować chatbota AI z identyfikatorem produktu

څنګه د محصول پیژندونکی AI چیټ بوټ جوړ کړئ

Jinsi ya Kuunda Chatbot ya Kitambulisho cha Bidhaa

தயாரிப்பு அடையாளங்காட்டி AI சாட்போட்டை எவ்வாறு உருவாக்குவது

Як створити чат-бот ШІ з ідентифікатором продукту

(1) Feng Liang, The University of Texas at Austin and Work partially done during an internship at Meta GenAI (Email: );
(2) Bichen Wu, Meta GenAI and Corresponding author;
(3) Jialiang Wang, Meta GenAI;
(4) Licheng Yu, Meta GenAI;
(5) Kunpeng Li, Meta GenAI;
(6) Yinan Zhao, Meta GenAI;
(7) Ishan Misra, Meta GenAI;
(8) Jia-Bin Huang, Meta GenAI;
(9) Peizhao Zhang, Meta GenAI (Email: );
(10) Peter Vajda, Meta GenAI (Email: );
(11) Diana Marculescu, The University of Texas at Austin (Email: ). Table of Links Abstract and Introduction
2. Related Work
3. Preliminary
4. FlowVid
4.1. Inflating image U-Net to accommodate video
4.2. Training with joint spatial-temporal conditions
4.3. Generation: edit the first frame then propagate


Experiments


5.1. Settings
5.2. Qualitative results
5.3. Quantitative results
5.4. Ablation study and 5.5. Limitations
Conclusion, Acknowledgments and References
A. Webpage Demo and B. Quantitative comparisons 5.3. Quantitative results User study We conducted a human evaluation to compare our method with three notable works: CoDeF [32], Rerender [49], and TokenFlow [13]. The user study involves 25 DAVIS videos and 115 manually designed prompts. Participants are shown four videos and asked to identify which one has the best quality, considering both temporal consistency and text alignment. The results, including the average preference rate and standard deviation from five participants for all methods, are detailed in Table 1. Our method achieved a preference rate of 45.7%, outperforming CoDeF (3.5%), Rerender (10.2%), and TokenFlow (40.4%). During the evaluation, we observed that CoDeF struggles with significant motion in videos. The blurry constructed canonical images would always lead to unsatisfactory results. Rerender occasionally experiences color shifts and bright flickering. TokenFlow sometimes fails to sufficiently alter the video according to the prompt, resulting in an output similar to the original video. Pipeline runtime We also compare runtime efficiency with existing methods in Table 1. Video lengths can vary, resulting in different processing times. Here, we use a video containing 120 frames (4 seconds video with FPS of 30). The resolution is set to 512 × 512. Both our FlowVid model and Rerender [49] use a key frame interval of 4. We generate 31 keyframes by applying autoregressive evaluation twice, followed by RIFE [21] for interpolating the non-key frames. The total runtime, including image processing, model operation, and frame interpolation, is approximately 1.5 minutes. This is significantly faster than CoDeF (4.6 minutes), Rerender (10.8 minutes) and TokenFlow (15.8 minutes), being 3.1×, 7.2×, and 10.5 × faster, respectively. CoDeF requires per-video optimization to construct the canonical image. While Rerender adopts a sequential method, generating each frame one after the other, our model utilizes batch processing, allowing for more efficient handling of multiple frames simultaneously. In the case of TokenFlow, it requires a large number of DDIM inversion steps (typically around 500) for all frames to obtain the inverted latent, which is a resource-intensive process. We further report the runtime breakdown (Figure 10) in the Appendix. This paper is available on arxiv under CC 4.0 license. (1) Feng Liang, The University of Texas at Austin and Work partially done during an internship at Meta GenAI (Email: ); (2) Bichen Wu, Meta GenAI and Corresponding author; (3) Jialiang Wang, Meta GenAI; (4) Licheng Yu, Meta GenAI; (5) Kunpeng Li, Meta GenAI; (6) Yinan Zhao, Meta GenAI; (7) Ishan Misra, Meta GenAI; (8) Jia-Bin Huang, Meta GenAI; (9) Peizhao Zhang, Meta GenAI (Email: ); (10) Peter Vajda, Meta GenAI (Email: ); (11) Diana Marculescu, The University of Texas at Austin (Email: ). (1) Feng Liang, The University of Texas at Austin and Work partially done during an internship at Meta GenAI (Email: ); (2) Bichen Wu, Meta GenAI and Corresponding author; (3) Jialiang Wang, Meta GenAI; (4) Licheng Yu, Meta GenAI; (5) Kunpeng Li, Meta GenAI; (6) Yinan Zhao, Meta GenAI; (7) Ishan Misra, Meta GenAI; (8) Jia-Bin Huang, Meta GenAI; (9) Peizhao Zhang, Meta GenAI (Email: ); (10) Peter Vajda, Meta GenAI (Email: ); (11) Diana Marculescu, The University of Texas at Austin (Email: ). Table of Links Abstract and Introduction 2. Related Work 3. Preliminary 4. FlowVid 4.1. Inflating image U-Net to accommodate video 4.2. Training with joint spatial-temporal conditions 4.3. Generation: edit the first frame then propagate Experiments 5.1. Settings 5.2. Qualitative results 5.3. Quantitative results 5.4. Ablation study and 5.5. Limitations Conclusion, Acknowledgments and References A. Webpage Demo and B. Quantitative comparisons Abstract and Introduction Abstract and Introduction 2. Related Work 2. Related Work 3. Preliminary 3. Preliminary 4. FlowVid 4. FlowVid 4 .1. Inflating image U-Net to accommodate video .1. Inflating image U-Net to accommodate video 4.2. Training with joint spatial-temporal conditions 4.2. Training with joint spatial-temporal conditions 4.3. Generation: edit the first frame then propagate 4.3. Generation: edit the first frame then propagate Experiments Experiments Experiments 5.1. Settings 5.1. Settings 5.2. Qualitative results 5.2. Qualitative results 5.3. Quantitative results 5.3. Quantitative results 5.4. Ablation study and 5.5. Limitations 5.4. Ablation study and 5.5. Limitations Conclusion, Acknowledgments and References Conclusion, Acknowledgments and References A. Webpage Demo and B. Quantitative comparisons A. Webpage Demo and B. Quantitative comparisons 5.3. Quantitative results User study We conducted a human evaluation to compare our method with three notable works: CoDeF [32], Rerender [49], and TokenFlow [13]. The user study involves 25 DAVIS videos and 115 manually designed prompts. Participants are shown four videos and asked to identify which one has the best quality, considering both temporal consistency and text alignment. The results, including the average preference rate and standard deviation from five participants for all methods, are detailed in Table 1. Our method achieved a preference rate of 45.7%, outperforming CoDeF (3.5%), Rerender (10.2%), and TokenFlow (40.4%). During the evaluation, we observed that CoDeF struggles with significant motion in videos. The blurry constructed canonical images would always lead to unsatisfactory results. Rerender occasionally experiences color shifts and bright flickering. TokenFlow sometimes fails to sufficiently alter the video according to the prompt, resulting in an output similar to the original video. User study Pipeline runtime We also compare runtime efficiency with existing methods in Table 1. Video lengths can vary, resulting in different processing times. Here, we use a video containing 120 frames (4 seconds video with FPS of 30). The resolution is set to 512 × 512. Both our FlowVid model and Rerender [49] use a key frame interval of 4. We generate 31 keyframes by applying autoregressive evaluation twice, followed by RIFE [21] for interpolating the non-key frames. Pipeline runtime The total runtime, including image processing, model operation, and frame interpolation, is approximately 1.5 minutes. This is significantly faster than CoDeF (4.6 minutes), Rerender (10.8 minutes) and TokenFlow (15.8 minutes), being 3.1×, 7.2×, and 10.5 × faster, respectively. CoDeF requires per-video optimization to construct the canonical image. While Rerender adopts a sequential method, generating each frame one after the other, our model utilizes batch processing, allowing for more efficient handling of multiple frames simultaneously. In the case of TokenFlow, it requires a large number of DDIM inversion steps (typically around 500) for all frames to obtain the inverted latent, which is a resource-intensive process. We further report the runtime breakdown (Figure 10) in the Appendix. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis: Quantitative result

(1) Feng Liang, The University of Texas at Austin and Work partially done during an internship at Meta GenAI (Email: );
(2) Bichen Wu, Meta GenAI and Corresponding author;
(3) Jialiang Wang, Meta GenAI;
(4) Licheng Yu, Meta GenAI;
(5) Kunpeng Li, Meta GenAI;
(6) Yinan Zhao, Meta GenAI;
(7) Ishan Misra, Meta GenAI;
(8) Jia-Bin Huang, Meta GenAI;
(9) Peizhao Zhang, Meta GenAI (Email: );
(10) Peter Vajda, Meta GenAI (Email: );
(11) Diana Marculescu, The University of Texas at Austin (Email: ). Table of Links Abstract and Introduction
2. Related Work
3. Preliminary
4. FlowVid
4.1. Inflating image U-Net to accommodate video
4.2. Training with joint spatial-temporal conditions
4.3. Generation: edit the first frame then propagate


Experiments


5.1. Settings
5.2. Qualitative results
5.3. Quantitative results
5.4. Ablation study and 5.5. Limitations
Conclusion, Acknowledgments and References
A. Webpage Demo and B. Quantitative comparisons 5.2. Qualitative results In Figure 5, we qualitatively compare our method with several representative approaches. Starting with a per-frame baseline directly applying I2I models, ControlNet, to each frame. Despite using a fixed random seed, this baseline often results in noticeable flickering, such as in the man’s clothing and the tiger’s fur. CoDeF [32] produces outputs with significant blurriness when motion is big in input video, evident in areas like the man’s hands and the tiger’s face. Rerender [49] often fails to capture large motions, such as the movement of paddles in the left example. Also, the color of the edited tiger’s legs tends to blend in with the background. TokenFlow [13] occasionally struggles to follow the prompt, such as transforming the man into a pirate in the left example. It also erroneously depicts the tiger with two legs for the first frame in the right example, leading to flickering in the output video. In contrast, our method stands out in terms of editing capabilities and overall video quality, demonstrating superior performance over these methods. We highly encourage readers to refer to more video comparisons in our supplementary videos. This paper is available on arxiv under CC 4.0 license. (1) Feng Liang, The University of Texas at Austin and Work partially done during an internship at Meta GenAI (Email: ); (2) Bichen Wu, Meta GenAI and Corresponding author; (3) Jialiang Wang, Meta GenAI; (4) Licheng Yu, Meta GenAI; (5) Kunpeng Li, Meta GenAI; (6) Yinan Zhao, Meta GenAI; (7) Ishan Misra, Meta GenAI; (8) Jia-Bin Huang, Meta GenAI; (9) Peizhao Zhang, Meta GenAI (Email: ); (10) Peter Vajda, Meta GenAI (Email: ); (11) Diana Marculescu, The University of Texas at Austin (Email: ). (1) Feng Liang, The University of Texas at Austin and Work partially done during an internship at Meta GenAI (Email: ); (2) Bichen Wu, Meta GenAI and Corresponding author; (3) Jialiang Wang, Meta GenAI; (4) Licheng Yu, Meta GenAI; (5) Kunpeng Li, Meta GenAI; (6) Yinan Zhao, Meta GenAI; (7) Ishan Misra, Meta GenAI; (8) Jia-Bin Huang, Meta GenAI; (9) Peizhao Zhang, Meta GenAI (Email: ); (10) Peter Vajda, Meta GenAI (Email: ); (11) Diana Marculescu, The University of Texas at Austin (Email: ). Table of Links Abstract and Introduction 2. Related Work 3. Preliminary 4. FlowVid 4.1. Inflating image U-Net to accommodate video 4.2. Training with joint spatial-temporal conditions 4.3. Generation: edit the first frame then propagate Experiments 5.1. Settings 5.2. Qualitative results 5.3. Quantitative results 5.4. Ablation study and 5.5. Limitations Conclusion, Acknowledgments and References A. Webpage Demo and B. Quantitative comparisons Abstract and Introduction Abstract and Introduction 2. Related Work 2. Related Work 3. Preliminary 3. Preliminary 4. FlowVid 4. FlowVid 4 .1. Inflating image U-Net to accommodate video .1. Inflating image U-Net to accommodate video 4.2. Training with joint spatial-temporal conditions 4.2. Training with joint spatial-temporal conditions 4.3. Generation: edit the first frame then propagate 4.3. Generation: edit the first frame then propagate Experiments Experiments Experiments 5.1. Settings 5.1. Settings 5.2. Qualitative results 5.2. Qualitative results 5.3. Quantitative results 5.3. Quantitative results 5.4. Ablation study and 5.5. Limitations 5.4. Ablation study and 5.5. Limitations Conclusion, Acknowledgments and References Conclusion, Acknowledgments and References A. Webpage Demo and B. Quantitative comparisons A. Webpage Demo and B. Quantitative comparisons 5.2. Qualitative results In Figure 5, we qualitatively compare our method with several representative approaches. Starting with a per-frame baseline directly applying I2I models, ControlNet, to each frame. Despite using a fixed random seed, this baseline often results in noticeable flickering, such as in the man’s clothing and the tiger’s fur. CoDeF [32] produces outputs with significant blurriness when motion is big in input video, evident in areas like the man’s hands and the tiger’s face. Rerender [49] often fails to capture large motions, such as the movement of paddles in the left example. Also, the color of the edited tiger’s legs tends to blend in with the background. TokenFlow [13] occasionally struggles to follow the prompt, such as transforming the man into a pirate in the left example. It also erroneously depicts the tiger with two legs for the first frame in the right example, leading to flickering in the output video. In contrast, our method stands out in terms of editing capabilities and overall video quality, demonstrating superior performance over these methods. We highly encourage readers to refer to more video comparisons in our supplementary videos. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis: Qualitative Results

Stories, News & Mentions about FLIKE

Flike’s Stories on HackerNoon

Flike's News & Mentions