ChatGPT 완전 미쳤다… AI 역사채널 자동제작｜구글 플로우 0크레딧 조건

0:00This video is one I created fully automatically with almost no additional credit costs. Let's watch just the first 30 seconds. In 612 BC, Nineveh collapsed. But beneath the ruins, something survived longer than swords. It was neither gold nor jewels, but clay tablets etched with writing. These were books made of earth instead of paper. Scribes pressed the tips of reeds into wet clay to preserve human knowledge. Every time the reed tip touched the clay, a small wedge-shaped mark was formed. From medicine to the movements of the stars. Even prayers

0:38and ancient stories. What do you think? The images move, the story is narrated, and the subtitles sync with the speech. The history documentary I made earlier is 15 minutes long, and this video you just saw is a sample where I recombined the images and audio. All I did was explain what kind of video I wanted and provided direction while watching the results. The production work—writing the script, preparing images and voices , editing, and syncing subtitles—was all handled by AI. I make the judgments , like choosing the topic and verifying

1:04that the completed content is correct. However, it didn't cost this little from the beginning. At first, I just made it with the hope that the results would turn out well. But then, when I calculated the credits used, it became too burdensome to continue in the same way. So I changed the video production method, switched the voice actor, and kept the parts I liked. Today, I'll show you that process along with the actual results. What I used at first, why I changed it, and how I arrived at my current production method? Once you see it, you'll get a sense of how the videos you want to make should be

1:35transformed. The first thing I made was a historical documentary about the ancient library of Nineveh. I wanted to show the world of ancient records through images and stories. I wanted a video where viewers could follow the story without me having to appear in every scene. At that time, the moving images were made with Google's Gemini 1.5 Flash. For the voice actor, I used ElevenLabs 'Arthur voice from Speechify . Let me show you the results from back then for a moment. Please take a look

2:02at the visuals and the voice together. In 612 BC, Nineveh fell. In the palace, the heart of the empire, broken clay tablets remained. In a city that the army could not protect, something unexpected survived. It felt like this. The atmosphere of the ancient city came through, and the voice matched the documentary style well. At first, I thought this was worth continuing to make. But a 15- minute video contains many scenes. It doesn't end just because one scene came out well. You make the next scene, change it if it doesn't fit the story, and remake it if there are any awkward parts. The voice also needs to be

2:40re-read if the script is revised. I asked for these accumulating costs to be calculated. I wanted to see how much was spent on the video and the voiceover separately. There is only one finished video, but costs can arise multiple times during production. The accounts I used are ElevenLabs and Google AI Ultra. I should start by saying that both are paid subscriptions . I selected the voice within ElevenLabs, and the video was created separately in Google Veo. The credits for the two services are in different units. For the first video, the cost was calculated at about 2,700 won for video, 7,700 won for voice, and 28,500 won including regenerations. This is an

3:19estimate based on converting used credits to won, not the actual amount charged to my card. Monthly subscription fees are separate, and the actual payment amount varies depending on whether the credits were included in the subscription or purchased additionally. For ElevenLabs, I calculated it assuming 20 credits per dollar at an exchange rate of 1,400 won . For Veo, I used the previously calculated standard of 14.4 won per credit. Getting good results is welcome , but if I worry about costs every time I make the next episode, it’s hard to keep doing it for long. The documentary flows calmly, but the credits were moving quite diligently. So, for the

3:52second attempt, I decided to try and reduce the video production costs. I changed the subject to the House of Wisdom in Baghdad and requested a slightly different art style. I initially thought about trying Veo 3.1 Fast, but changed it to Light to make it even lighter. The version I actually completed is Veo 3.1 Light. The setting is Baghdad. It is a city that collected books from distant lands and asked new questions. Have you heard the name "House of Wisdom"? It is a famous story about the world's

4:22knowledge gathering under one roof. This one feels different, doesn't it? Because the art style changed, the atmosphere of the history documentary changed as well. There were parts of this version that I liked, too. What I wanted to see here was whether it was sufficient to convey the story I wanted . Not every scene in history needs to move like an action movie scene. A scene can come to life just by having a person reading a book, a slight hand

4:46movement, or light entering the space. If you choose the movement that suits the purpose of the video, your options change as well. I am looking at whether a scene requires major changes in expressions and actions, or if it just needs subtle movements while maintaining the atmosphere of the image . I found that combination to suit the documentary I wanted. For Baghdad, the video was about 8,100 won and the voiceover was about 5,700 won. The total converted cost, including regenerations, was about 13,900 won. On the Ultra account I used, standard Lite was 5 credits per generation. Video costs were reduced, but voiceover costs still remained separately. After

5:20changing the video, the voiceover cost caught my eye again. At the time, I was using ElevenLabs' Arthur for the voice. I liked the voice, but as I kept creating long scripts and revisions, it became more burdensome than I expected. So I asked again. Is there a way to use this voice more cheaply? Can I achieve

5:38this same vibe through other methods? That's how I found Google Cloud's Text-to-Speech. It's called TTS for short, and it's a feature that reads text aloud when you input it. Ultimately, I switched to using voices from this service. Take a listen. Some read it, and some copied it again. The kingdom is gone, but we read its

6:01fragments again. How is it? Even using the free tier, the quality is good enough. I felt it was fine for delivering these calm historical stories. It has a different feel from the voice I used before. You can listen to the difference and choose . You just need to see if it fits the mood of the video you're making, if the Korean is comfortable to listen to, and if it's easy on the ears for a long time. When we say "free" here, you have to look at the conditions together. The key to the combination I found is Veo 3.1 Lite Low Priority. The selection

6:29options are different from just "Lite." Google AI Ultra subscribers or family plan managers have 0 generation credits when making videos with this model. This benefit doesn't apply equally to regular family members or other plans. You must check the exact model name and the 0 credit indicator in the flow. The existing subscription fee is separate, but credits aren't deducted when creating additional moving scenes. I changed the voiceover to the Google Cloud Chirp 3HD Caron voice. You need to link a billing account to your cloud project and enable the Text-to-Speech API. It doesn't mean you can use it for free without registering a card. The voice model I chose has a free tier of

7:06up to 1 million characters per month. As long as this limit remains, the additional synthesis fee for narration is also zero. If you exceed the free tier, a fee of $ 30 per 1 million characters is applied. So, if you meet the necessary subscription and free limit conditions, both the moving video and the additional narration generation costs become zero. I found this part truly amazing. The result created with those conditions is the history documentary you just saw. The Chris voice reading this explanation is from ElevenLabs, so the production cost is different from the Caron voice you heard earlier. So, before starting, I made sure to check the conditions

7:37available for my current account. You shouldn't assume that just because it was free before, it will be the same next time. I wanted to balance visual quality with the cost of producing content consistently. At first, I made it using the Omni and Arthur voices. Next, I switched the video over to actors, and finally, I changed the voice to Google Cloud. Then, I applied a cost-effective video generation method to the original art style I liked. The result of that process became the standard for the video I showed you at the beginning. What I want to convey to you here is that you don't have to stick to your initial

8:08combination to the end. If you like the visuals but find the voice burdensome, you can try changing the voice. If the voice is fine but the art style doesn't fit, you can try changing the images. By separating the roles like this, you can make edits while keeping the parts you like. You don't need to rebuild a house just because a lightbulb inside is too bright. Knowing exactly what to change reduces the need to redo work that was already done well. So, how did I manage all these processes? The platform where I directed the entire project was Codex. I explained the direction of the video I wanted to

8:40create and requested the necessary tasks to follow. The first thing I decided was what kind of video to make. It’s about deciding what kind of feeling you want to leave the audience with after telling them your story. In my case, it was a historical documentary that calmly shares old knowledge and records. Even when describing the same library, there are many different ways to do it. You could focus on the architecture, or talk about how the books survived. I set this direction and entrusted it to prepare the script and scenes accordingly. At first, it's easy for anyone to start by choosing tools. "I heard this drawing tool is good.""I

9:13heard that voice sounds natural." While you're checking out new features, you realize you haven't actually decided what video to make. It’s like having a shopping cart full of cooking utensils but no menu for dinner. Once you decide on the desired result first, you'll have criteria for choosing your tools. Because then you can judge if a feature is necessary for your current video. My decision to swap models and voices was also a choice made within those criteria. The second step was to ensure the results from the previous task were used for the next one. You have to determine what stories are in the script to know what images you need

9:45. You need to have the actual voiceover to decide how long to show each scene. Subtitles must match that audio so the viewers can follow along comfortably. When this sequence is organized, you spend less time finishing tasks separately and then trying to sync them up later. Even if the script, voice, and visuals are created well individually, they can feel awkward when put together. The narrator might be talking about something else while

10:07the previous scene is still on screen. I wanted to entrust even that connectivity to someone else. I didn't want to keep copying the script, pasting it into other windows, downloading files, and moving them to an editor. I connected the accessible tools and work files so that results from one step could be used in the next . You need to prepare an environment where you can log in to the service and use the necessary features. There were times when I had to check account settings while working. This is a method of delegating repetitive production tasks within the scope of that prepared environment. During the production process, the results I saw

10:40were also important. For example, a voice might read words accurately but still feel somewhat rushed. If they don't pause where they should, it’s hard for the listener to follow the story. I actually had parts like that, so I asked them to change where the pauses were. I organized the sentences to be read and refined the actual spacing of the voice. Even though it was the same content, it sounded much more comfortable. For the subtitles, I

11:03asked them to show numbers as digits. Sometimes it’s more natural to say numbers in words, but it’s easier to see them as digits on screen. I organized the expressions for the eyes and the expressions for the ears separately. I also asked for the subtitles to appear in sync with what I was actually hearing. If a sentence is long, I break it into readable units and ensure it doesn't disappear before the speech ends. I explained these standards and pointed out awkward parts while reviewing the results. I didn't use many technical production terms during this process. "The words are too close together here.""This phrase is difficult to read at once.""I would

11:35like the scene transitions to match the story." It was more like that. These are judgments you can easily make from the perspective of a viewer. If you know which parts are uncomfortable, you can explain what needs to be changed. You don't have to know how to do everything yourself to have an opinion on the results. I also checked if the necessary scenes flowed naturally on screen. If you're explaining someone writing a book, it's easier to understand if you show their hands or the book. When the story moves to a different place, it's good to show the appearance of the new location. The goal is to help the viewer follow the

12:06story while watching the screen. The third point was to save the standards I liked so I could use them next time. I checked the visuals I liked and the revised voice and subtitles together in a short sample. After seeing that, I developed a standard of what I considered acceptable. And I organized those standards so they could be

12:23applied to create the entire story. Having a short reference video makes explanations more concrete. "Please make it calm.""Please make it natural." These phrases mean different things to different people. It is easier to share a standard by showing a short result and saying, "This is the tone of voice and this is the level of movement." Just like how you can choose clothes you like even if you don't know how to make them yourself. Rather than trying to give the perfect explanation from the start, just point out what you liked after seeing the result. After making a 15-minute video and a short sample, I wanted to carry over the

12:55choices I made to the next subject. So, I asked to have my instructions and revisions organized into a production template. From art style, voice, and subtitles to cost conditions and final check items. I also left a reference video. I even named it "Documentary Template" so it would be easy for me to identify. It’s so I can call up these standards again when making something else next time. I made it so I wouldn’t have to decide everything, from the voice to the subtitles, every single time. Using a template doesn't

13:22mean the story is the same every time. When the subject changes, the characters and settings change, and the story to be told changes as well. Within that, you can maintain the tone of the voice, the visual style, and the reading experience of the subtitles. And the revisions I made become the starting point for the next production. If the subtitles were too long this time, I leave a rule to shorten them, and if the art styles were mixed, I leave a rule to use reference images consistently. It would be a waste to explain from scratch the things we already solved last time. This is why I am interested in automation. Because

13:52while producing something once, you can leave behind not just the video file, but also methods to assist with the next task. When you have your next idea , you are already a bit more prepared to start again. By now, you might have this thought. If everyone uses the same tools, won't everyone end up making similar videos? That is why I believe choosing the right story becomes even more important. Even for the same library, the building could be the protagonist, or the person who worked there could be the protagonist. Someone might be curious about the contents of a book, while someone else might be curious about the process of how that

14:22book disappeared and was discovered. Starting with which question and where to resolve the viewer's curiosity is a choice for the creator to make. The more you delegate repetitive tasks, the more you can focus on these choices. You take the attention you used for moving files and naming them and use it to check if the current story is actually interesting. If you are dealing with history or knowledge, you

14:40also need to verify the content. Because if the visuals are great and the voice is natural, it can feel like the explanation itself is also correct. So, you must cross-check important years, figures, and events with reliable sources. You need to verify the impressive visuals and the accurate descriptions separately. And simply finishing a video doesn’t guarantee views or revenue. You also have to ensure the topic interests viewers and is explained in a way that keeps them watching until the end. I believe that reducing the burden of production gives you the room to consistently experiment . It’s overwhelming if you think you have to make every video for your

15:11channel from the start. Start by deciding on just one story you want to tell. Look at the short result of that story, pick the parts you like, and fix what feels awkward. Then, you can apply the standards you’ve found to the full-length version. The video I showed you at the beginning went through that exact process. I started with an AI voice, and as I checked costs, I swapped out the visuals and the narrator. Even so, I tried to preserve the imagery and mood that I liked. I kept the answers I found during production so I could use them again later. I hope you don’t just watch

15:41this, but also take a moment to think about what kind of video you want to make. It could be a story about history , books, or even an explanation of something you know well. Once you decide who you want to talk to and what you want to say, learning production methods becomes much more practical. That’s why I’ve prepared a free lecture to teach you the secret to making these videos with AI. I provide the detailed prompts used in making the video to my course students. I’d love for those who want to create their first video with their own material to join. There are three steps to sign up

16:14for the free lecture. First, please hit the like button on this video. Second, leave a comment with your thoughts on this video. Third, go to the link in the pinned comment and complete the class registration. I will send the lecture link on the day of the session to those who have completed all three: liking, commenting with thoughts, and registering for the class. The lecture schedule will be sent via email, so please check the email address you used to register. After leaving a comment, please make sure to complete your class registration via the pinned comment link. Just like the video I showed you today, I hope you turn the stories you

16:45want to tell into actual results. Let’s make that start together in the free lecture. This has been Nomad Chris .