Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

#SakuraWars2 DevLog: Dealing with Images This game contains a fair bit of text within images. So for a translation patch, the images need to be extracted, redone, and patched.

NSteam 🇵🇸

1,408 subscribers

19,529 views • 1 year ago •via X (Twitter)

Gaming Science & Technology #SakuraWars2

Anya Rossi• Live Now

Private livecam show

10 Comments

NSteam 🇵🇸1 year ago

This includes font sheets, all battle UI, chapter intro screens, battle screens, boss banners, minigame images, theater room names, system menus, menu backgrounds, loading screens, splash screen, game title screen, and other misc images. Currently amounting to 1,735 images!

NSteam 🇵🇸1 year ago

For some of these, like the room names in the theater & names appearing within battle UI, there were hundreds of images. To do them by hand was tedious so I wrote a tool to auto-generate them given the translated names in a text file. Some of the battle UI was done by hand though

NSteam 🇵🇸1 year ago

We got the help of some very talented artists to help redo some of the more complicated images like these boss images, chapter screens, and minigame images.

NSteam 🇵🇸1 year ago

Most images are stored as raw vdp1 format image data. Others are tiled images displayed in vdp2. These are easy to extract using tools I wrote. Others are compressed using PRS compression which was a standard across many Sega titles. These are a bit more complicated.

NSteam 🇵🇸1 year ago

Some however are encoded using some bizarre encoding algorithm. It doesn't compress the image, just obfuscates it. The Kinematron, some minigames, and credits all use this. And these are a pain.

NSteam 🇵🇸1 year ago

To extract these, the encoded data needs to be found, the encoding algorithm needs to be figured out(which Ghidra helps with) and then it needs to be reverse engineered so that I can re-encode the translated images and this part was pretty tricky.

NSteam 🇵🇸1 year ago

Above image is full of artifacts unless it is decoded. Here is one of the images properly decoded. What makes it worse is that there are multiple encoding schemes. The kinematron & minigames use one, credits use another. I call this algorithm the "Kinematron Encoding Algorithm"

NSteam 🇵🇸1 year ago

I have yet to reverse engineer the credits algorithm. Not sure if it will make it into the final patch. Maybe! But for now, everything else we have encountered has been extracted, reworked, and patched =)

burntends | Macross Do You Remember Love?1 year ago

Always great to see more news

Norbjunior1 year ago

@burntends2 Super cool to learn how some of this works!! Thank you for all the hard work!!!!

Related Videos

Euclid's first images are here ✨ These five images illustrate Euclid's full potential; never before has a telescope been able to create such razor-sharp images across such a large patch of the sky, and looking so far into the distant Universe. 👉

Euclid's first images are here ✨ These five images illustrate Euclid's full potential; never before has a telescope been able to create such razor-sharp images across such a large patch of the sky, and looking so far into the distant Universe. 👉

European Space Agency

312,724 views • 2 years ago

✨ I made my first video game with ChatGPT: 1) ChatGPT generates a text-based adventure game with DALL-E 3 generating images for it 2) Every time you play the game is different because it generates the story and images live 3) The images from DALL-E are sent to Runway which turns images into video 4) The text is sent to ElevenLabs which turns the text adventure into a pirate narrator voice 5) It's merged into a video 6) Interactive buttons are overlayed The game is called: 🐒🏝️🇳🇱The Secret of Monkey Island: Amsterdam (unofficial) And you can play it here: (video + TTS + buttons doesn't work auto yet, for now manual but text + img works, I'm building an interface for it now)

✨ I made my first video game with ChatGPT: 1) ChatGPT generates a text-based adventure game with DALL-E 3 generating images for it 2) Every time you play the game is different because it generates the story and images live 3) The images from DALL-E are sent to Runway which turns images into video 4) The text is sent to ElevenLabs which turns the text adventure into a pirate narrator voice 5) It's merged into a video 6) Interactive buttons are overlayed The game is called: 🐒🏝️🇳🇱The Secret of Monkey Island: Amsterdam (unofficial) And you can play it here: (video + TTS + buttons doesn't work auto yet, for now manual but text + img works, I'm building an interface for it now)

@levelsio

2,725,149 views • 2 years ago

I needed an app to export markdown from images and PDFs, so I built one using the best model on the market: Mistral OCR. It can convert images, PDFs, and even extract tables and images, keeping them contextually within the markdown. All you need is a Mistral AI API.

I needed an app to export markdown from images and PDFs, so I built one using the best model on the market: Mistral OCR. It can convert images, PDFs, and even extract tables and images, keeping them contextually within the markdown. All you need is a Mistral AI API.

Pietro Schirano

48,786 views • 1 year ago

Within two years, you can create a full game like this from scratch in less than a week. In the meantime, here's the prompt to create the images for a game like this with Midjourney 👇🏼

Within two years, you can create a full game like this from scratch in less than a week. In the meantime, here's the prompt to create the images for a game like this with Midjourney 👇🏼

PJ Ace

354,218 views • 1 year ago

A demo of a document translation feature coming soon to the Khaya AI web app. Text can be extracted from images (OCR), PDF, and docx. Built entirely by our National Service employee Nana Akwasi Asare Next time some 🤡 says Ghanaian kids don't have talent, make u slap em give me 🤣

A demo of a document translation feature coming soon to the Khaya AI web app. Text can be extracted from images (OCR), PDF, and docx. Built entirely by our National Service employee Nana Akwasi Asare Next time some 🤡 says Ghanaian kids don't have talent, make u slap em give me 🤣

Paul Azunre

18,456 views • 1 year ago

The LVM (large vision model) revolution is coming a little after the LLM (large language model) one, and will transform how we process images. But there’s an important difference between LVMs and LLMs: - Internet text is similar enough to proprietary text documents that an LLM trained on internet text can understand your documents. - But internet images – such as Instagram pictures – contain a lot of pictures of people, pets, landmarks, and everyday objects. Many practical vision applications (manufacturing, aerial imagery, life sciences, etc.) use images that look nothing like most internet images. So a generic LVM trained on internet images fares poorly at picking out the most salient features of images in many specialized domains. That’s why domain specific LVMs – ones adapted to images of a particular domain (such as semiconductor manufacturing, or pathology) – do much better. At LandingAI , by using ~100K unlabeled images to adapt an LVM to a specific domain, we see significantly improved results, for example where only 10-30% as much labeled data is now needed to achieve a certain level of performance. For companies with large sets of images that look nothing like internet images, I think domain specific LVMs can be a way to unlock considerable value from their data. Dan Maloney and I share more details in the video.

The LVM (large vision model) revolution is coming a little after the LLM (large language model) one, and will transform how we process images. But there’s an important difference between LVMs and LLMs: - Internet text is similar enough to proprietary text documents that an LLM trained on internet text can understand your documents. - But internet images – such as Instagram pictures – contain a lot of pictures of people, pets, landmarks, and everyday objects. Many practical vision applications (manufacturing, aerial imagery, life sciences, etc.) use images that look nothing like most internet images. So a generic LVM trained on internet images fares poorly at picking out the most salient features of images in many specialized domains. That’s why domain specific LVMs – ones adapted to images of a particular domain (such as semiconductor manufacturing, or pathology) – do much better. At LandingAI , by using ~100K unlabeled images to adapt an LVM to a specific domain, we see significantly improved results, for example where only 10-30% as much labeled data is now needed to achieve a certain level of performance. For companies with large sets of images that look nothing like internet images, I think domain specific LVMs can be a way to unlock considerable value from their data. Dan Maloney and I share more details in the video.

Andrew Ng

685,054 views • 2 years ago

Today, Bullhorn Betty accidentally exposed the "enhanced video" fraud during her live stream. She revealed a 13-page folder containing images and background layers used to fabricate the so-called "enhanced" image. Among these images was a frame from the Idaho 4 murders, proof that these so-called enhancements are nothing more than manipulated fakes created for profit. For an image to be truly enhanced, only the actual source video can be used. However, the folder she revealed contains 13 images, including background color layers and external images that were not sourced from the original lights video. This is proof of deception. Bullhorn Betty #FlopAtLife #frauds

Today, Bullhorn Betty accidentally exposed the "enhanced video" fraud during her live stream. She revealed a 13-page folder containing images and background layers used to fabricate the so-called "enhanced" image. Among these images was a frame from the Idaho 4 murders, proof that these so-called enhancements are nothing more than manipulated fakes created for profit. For an image to be truly enhanced, only the actual source video can be used. However, the folder she revealed contains 13 images, including background color layers and external images that were not sourced from the original lights video. This is proof of deception. Bullhorn Betty #FlopAtLife #frauds

SF INVESTIGATES • STEVE FISCHER

38,882 views • 1 year ago

‼️ Warning… the video contains graphic images of a Wisconsin man in a cold lake. Not safe for Californians and career politicians.‼️

‼️ Warning… the video contains graphic images of a Wisconsin man in a cold lake. Not safe for Californians and career politicians.‼️

Eric Hovde

1,205,238 views • 2 years ago

Image generation getting better on Notion. Still need to make it faster, but you can see how you can generate images with prompts, auto-generate cover images and even edit images right inside a doc!

Image generation getting better on Notion. Still need to make it faster, but you can see how you can generate images with prompts, auto-generate cover images and even edit images right inside a doc!

Akshay Kothari

47,259 views • 6 months ago

You just don’t see quality tillage commercials like this anymore. Caution to no-tillers, this video contains images of a plow

You just don’t see quality tillage commercials like this anymore. Caution to no-tillers, this video contains images of a plow

Holly

115,137 views • 3 years ago

When McKay Coppins set out to report on the explosive growth of the sports-betting industry, his editors thought he should experience the phenomenon firsthand—so he spent an NFL season as an online gambler. His reporting took him to Las Vegas, where he wanted to see how betting online compared with gambling the old-fashioned way. Coppins needed a guide to show him around the casino. Atlantic staff writer Tom Nichols stepped in. Read more about their time in Vegas, and find Coppins’s April cover story, at the link. 📸: Michael Reaves / Getty Images; Harry How / Getty Images; Thearon W. Henderson / Getty Images; Nextrecord Archives / Getty Images; D. Corson / ClassicStock / Getty Images; B. Tanaka / Corbis Documentary / Getty Images; Keystone-France / Gamma-Keystone / Getty Images; Yoshikazu Tsuno / AFP / Getty Images; Bettmann / Getty Images; ullstein bild / Getty Images; EvgeniyShkolenko / Creatas Video+ / Getty Images; BBC Archive / BBC Creative/ Getty Images; Clive Mason / Getty Images; Bernard Annebicque / Sygma/Sygma / Getty Images; Hulton-Deutsch Collection / Corbis / Getty Images; Michael Montfor t /Michael Ochs Archives / Getty Images; Found Image Holdings / Corbis / Getty Images; George Rose / Getty Images.

When McKay Coppins set out to report on the explosive growth of the sports-betting industry, his editors thought he should experience the phenomenon firsthand—so he spent an NFL season as an online gambler. His reporting took him to Las Vegas, where he wanted to see how betting online compared with gambling the old-fashioned way. Coppins needed a guide to show him around the casino. Atlantic staff writer Tom Nichols stepped in. Read more about their time in Vegas, and find Coppins’s April cover story, at the link. 📸: Michael Reaves / Getty Images; Harry How / Getty Images; Thearon W. Henderson / Getty Images; Nextrecord Archives / Getty Images; D. Corson / ClassicStock / Getty Images; B. Tanaka / Corbis Documentary / Getty Images; Keystone-France / Gamma-Keystone / Getty Images; Yoshikazu Tsuno / AFP / Getty Images; Bettmann / Getty Images; ullstein bild / Getty Images; EvgeniyShkolenko / Creatas Video+ / Getty Images; BBC Archive / BBC Creative/ Getty Images; Clive Mason / Getty Images; Bernard Annebicque / Sygma/Sygma / Getty Images; Hulton-Deutsch Collection / Corbis / Getty Images; Michael Montfor t /Michael Ochs Archives / Getty Images; Found Image Holdings / Corbis / Getty Images; George Rose / Getty Images.

The Atlantic

14,692 views • 4 months ago

I need to annotate some images for training a computer vision model. There are many powerful annotation platforms available, but I want to keep my images local. I added a new section to my CV Streamlit app to quickly annotate images and train a YOLO model in a few clicks.

I need to annotate some images for training a computer vision model. There are many powerful annotation platforms available, but I want to keep my images local. I added a new section to my CV Streamlit app to quickly annotate images and train a YOLO model in a few clicks.

Marco Franzon

29,530 views • 7 months ago

🚨 ELON: GROK WILL BE ABLE TO UNDERSTAND IMAGES, VIDEOS AND BEHAVIORS “It does, it actually does understand images at this point. It certainly is intended to work in sync. It's intended to be what's called a multimodal model, which means understanding text, images, video, and audio.” Peterson: “Okay, if it understands video, will it start to understand behavior?” Elon: “Yeah.” Source: DW

🚨 ELON: GROK WILL BE ABLE TO UNDERSTAND IMAGES, VIDEOS AND BEHAVIORS “It does, it actually does understand images at this point. It certainly is intended to work in sync. It's intended to be what's called a multimodal model, which means understanding text, images, video, and audio.” Peterson: “Okay, if it understands video, will it start to understand behavior?” Elon: “Yeah.” Source: DW

Mario Nawfal

58,182 views • 2 years ago

When creating ingredient images, try to create stylistically consistent images. If you use input images that are too disparate, the output may lead to a cut scene. For example, when generating a reference image, try using a prompt like: “Create the next scene in this storyboard, now show a roadside diner with the same look and feel as the original image.”

When creating ingredient images, try to create stylistically consistent images. If you use input images that are too disparate, the output may lead to a cut scene. For example, when generating a reference image, try using a prompt like: “Create the next scene in this storyboard, now show a roadside diner with the same look and feel as the original image.”

Google AI

23,526 views • 10 months ago

If you're building a PDF RAG pipeline: Should you be using OCR and 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 methods, or just 𝗲𝗺𝗯𝗲𝗱 𝗶𝗺𝗮𝗴𝗲𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 using late interaction models? This paper says the answer might actually be 𝘣𝘰𝘵𝘩. My colleagues at Weaviate released IRPAPERS, a benchmark comparing 𝗶𝗺𝗮𝗴𝗲-𝗯𝗮𝘀𝗲𝗱 and 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 retrieval over 3,230 pages from 166 scientific papers. The setup: Take the same PDFs and process them two ways. For text, run OCR with GPT-4.1 and embed with Arctic 2.0 + BM25 hybrid search. For images, embed raw page images with ColModernVBERT multi-vector embeddings. Test both on 180 needle-in-the-haystack questions. 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝘀: Text edges out images at the top rank: 46% vs 43% Recall@1 But images match or exceed text at deeper recall: 93% vs 91% Recall@20 But text and image based methods actually fail on 𝘥𝘪𝘧𝘧𝘦𝘳𝘦𝘯𝘁 𝘲𝘶𝘦𝘳𝘪𝘦𝘴. At Recall@1: • 22 queries succeed with text but fail with images • 18 queries succeed with images but fail with text This complementarity is what makes 𝗠𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝗛𝘆𝗯𝗿𝗶𝗱 𝗦𝗲𝗮𝗿𝗰𝗵 work. By fusing scores from both text and image retrieval, they achieved: • 49% Recall@1 (beating either modality alone) • 81% Recall@5 • 95% Recall@20 More in the video below 🔽 Dataset: Paper: Code:

If you're building a PDF RAG pipeline: Should you be using OCR and 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 methods, or just 𝗲𝗺𝗯𝗲𝗱 𝗶𝗺𝗮𝗴𝗲𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 using late interaction models? This paper says the answer might actually be 𝘣𝘰𝘵𝘩. My colleagues at Weaviate released IRPAPERS, a benchmark comparing 𝗶𝗺𝗮𝗴𝗲-𝗯𝗮𝘀𝗲𝗱 and 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 retrieval over 3,230 pages from 166 scientific papers. The setup: Take the same PDFs and process them two ways. For text, run OCR with GPT-4.1 and embed with Arctic 2.0 + BM25 hybrid search. For images, embed raw page images with ColModernVBERT multi-vector embeddings. Test both on 180 needle-in-the-haystack questions. 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝘀: Text edges out images at the top rank: 46% vs 43% Recall@1 But images match or exceed text at deeper recall: 93% vs 91% Recall@20 But text and image based methods actually fail on 𝘥𝘪𝘧𝘧𝘦𝘳𝘦𝘯𝘁 𝘲𝘶𝘦𝘳𝘪𝘦𝘴. At Recall@1: • 22 queries succeed with text but fail with images • 18 queries succeed with images but fail with text This complementarity is what makes 𝗠𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝗛𝘆𝗯𝗿𝗶𝗱 𝗦𝗲𝗮𝗿𝗰𝗵 work. By fusing scores from both text and image retrieval, they achieved: • 49% Recall@1 (beating either modality alone) • 81% Recall@5 • 95% Recall@20 More in the video below 🔽 Dataset: Paper: Code:

Victoria Slocum

44,145 views • 5 months ago

Georgia's 2020 election had 1,726,651 missing original ballot images. You can NOT certify or audit votes without these images, many were destroyed or "lost." 70 Georgia counties could not produce original ballot images for the November 2020 election, 36 counties were missing 100% of all original ballot images. These images MUST be retained up to 22 months by law. These ballot images were deleted because it proved the massive amount of fraud, and they had to get rid of the evidence. 5,026,684 total cast ballots from the Secretary of State, 3,300,033 original ballot images received by VoterGA. The amount of "anomalies," impossible occurrences, and patterns of fraud, happening all throughout the country clearly shows a stolen election.

Georgia's 2020 election had 1,726,651 missing original ballot images. You can NOT certify or audit votes without these images, many were destroyed or "lost." 70 Georgia counties could not produce original ballot images for the November 2020 election, 36 counties were missing 100% of all original ballot images. These images MUST be retained up to 22 months by law. These ballot images were deleted because it proved the massive amount of fraud, and they had to get rid of the evidence. 5,026,684 total cast ballots from the Secretary of State, 3,300,033 original ballot images received by VoterGA. The amount of "anomalies," impossible occurrences, and patterns of fraud, happening all throughout the country clearly shows a stolen election.

The SCIF

95,822 views • 1 month ago

Collaborative Score Distillation for Consistent Visual Synthesis paper page: Generative priors of large-scale text-to-image diffusion models enable a wide range of new generation and editing applications on diverse visual modalities. However, when adapting these priors to complex visual modalities, often represented as multiple images (e.g., video), achieving consistency across a set of images is challenging. In this paper, we address this challenge with a novel method, Collaborative Score Distillation (CSD). CSD is based on the Stein Variational Gradient Descent (SVGD). Specifically, we propose to consider multiple samples as "particles" in the SVGD update and combine their score functions to distill generative priors over a set of images synchronously. Thus, CSD facilitates seamless integration of information across 2D images, leading to a consistent visual synthesis across multiple samples. We show the effectiveness of CSD in a variety of tasks, encompassing the visual editing of panorama images, videos, and 3D scenes. Our results underline the competency of CSD as a versatile method for enhancing inter-sample consistency, thereby broadening the applicability of text-to-image diffusion models.

Collaborative Score Distillation for Consistent Visual Synthesis paper page: Generative priors of large-scale text-to-image diffusion models enable a wide range of new generation and editing applications on diverse visual modalities. However, when adapting these priors to complex visual modalities, often represented as multiple images (e.g., video), achieving consistency across a set of images is challenging. In this paper, we address this challenge with a novel method, Collaborative Score Distillation (CSD). CSD is based on the Stein Variational Gradient Descent (SVGD). Specifically, we propose to consider multiple samples as "particles" in the SVGD update and combine their score functions to distill generative priors over a set of images synchronously. Thus, CSD facilitates seamless integration of information across 2D images, leading to a consistent visual synthesis across multiple samples. We show the effectiveness of CSD in a variety of tasks, encompassing the visual editing of panorama images, videos, and 3D scenes. Our results underline the competency of CSD as a versatile method for enhancing inter-sample consistency, thereby broadening the applicability of text-to-image diffusion models.

AK

33,513 views • 3 years ago