# ChatGPT erstellt Videos in Rekordzeit 🤯 lohnt sich GPT-6 Astra? 🎬

Channel: Felicia Simon
Video: https://www.youtube.com/watch?v=N9SVqDkk0WE
Duration: 19 min
Language: English
Words: 3873
Transcript page: https://viewrankai.com/tools/youtube-transcript/N9SVqDkk0WE

---

[0:00] With this single short prompt, Chat Be Work generated a finished explanatory video for me in just a few minutes, and with the same prompt, created an entire advertising campaign as well. And GPD6 Astra took over the direction. This is currently the most powerful AI model from Open AI and you can implement truly incredible things with it. But it is also the most expensive AI model from Open AI, which is responsible for burning through your tokens at a very rapid pace. That's why we're going to take a look together at when it's really worth it and how you can generate the best AI videos. As always, everything is explained simply, that's

[0:29] the plan. Let's go. So, another round. It's Monday again. Great that you're joining us. And today, we are creating AI videos with GPT6 Astra and are also taking the time to compare the results with cheaper AI models that are available from Opmi. We will create an advertisement, an AI film, and a completed explanatory video together to test exactly what it is that Astra is capable of doing effectively. In the process, you'll also learn what you should definitely consider when creating AI videos with Chat GPT and when the most expensive model is really worth it. GPD6 Astra is currently the most powerful AI model in CHGPT and

[1:03] takes over the management of your film production. And for that, you need another tool and that is HF AI. This video here is sponsored by Hixfield and I myself have been using Hixfield AI for a very, very long time. So, you have the idea, Chat BD takes over the direction and manages the film production and Hixfield generates the videos for you with the best video AI models. You will therefore need both a Chat GPT account and a Hixfield account ; connecting the two tools is quite simple if you open the plugins within Chat GPT and then proceed to enter Hixfield in the designated search field provided there. By selecting Install

[1:33] Plugin, you will be guided through the entire process and will be able to connect the two tools with one another successfully. And when you then start a new chat here, you can quite easily select Hickfield by using the plus sign and then proceed to enter your own individual prompt. I will select Chat GPT Work here, and now we want to create an advertising campaign together , but for that purpose, we first need to decide upon a product. In this specific case, we are going to generate a digital advertisement for a running shoe. And now we are conducting the first test to see whether the AI or

[1:56] Chat GPT actually makes a significant difference when it comes to generating these images. So, I selected Hixfield here in the chat by using the plus button and then I entered the following prompt: "Create a realistic product photo for a stylish, colorful running shoe." And here you can select the AI model in Chat GBT. The GPD6 Astra is the most powerful and expensive model, while the GPD6 Luna serves as the faster and more affordable option in the series; the Soul 6.1 is a high-quality model that is positioned to be more affordable than the Astra, yet costs more than the Luna. Here you can also see the prices in the overview

[2:28] . I ran the prompt through a sequence of three tests: once with GPD6 Astra, once with GPD 6.1 inches, and finally with GPD6 Luna. And these are the three results. Luna needed 32 seconds to complete the process and this image here is what resulted. The image was created with GPT Image 2.5 FLIR. So this is the image AI model which Chat GPT has decided to make use of at this particular point in time. GPT 6.1 1 inch took 52 seconds, and the image was also created using GPT Image 2.5 FLIR, while Astra took 56 seconds and utilized the exact same image model for

[2:58] the generation of that specific result. In that case, I would say that Astra is not really worth it for such a simple image generation. The prompt was extremely vague, and in order to achieve much better results, I would advise you to use reference images and provide more specific details, rather than just wasting all your tokens on GPT6 Astra. You can save money using Chat GPT, or you can also quite simply create your images directly within Chat GPT and thereby save yourself the costs that would otherwise arise for image generation in Hickfield. And now we come to the advertisement. To do this, we are going to take one of the images

[3:28] and drag it right into the chat. Select Hickfield again and add the prompt. Create an emotional advertising video for this stylish running shoe. Duration 30 seconds 16: 9. So another very, very short prompt, the image information about the duration and aspect ratio. And before I reveal the result to you, if you happen to like this video by then, I would truly appreciate your support with a thumbs up. Please consider subscribing to my channel, and do not forget to activate the notification bell as well, so that you will never miss out on any of the new videos that I upload for you every single Monday. Yes, I know, I repeat

[3:57] this in every single video, and believe me, I do not like these calls to action either, but so many people on YouTube unfortunately forget to leave a thumbs up for the video. And that is exactly why I am making this small request to you, and if you choose to support me here, I can continue to use my videos to help you understand AI and to utilize it efficiently. I have now entered this prompt here and have once selected GPD6 Luna with little effort, and another time I have selected GPD6 Astra with little effort. Luna took 52 seconds to finish, while Astra took 1 minute and 6 seconds. So Luna was

[4:26] faster, and we are now going to take a look at the results together. This is the ad that Luna created. And that is the Astra advertising video in a direct comparison. Sometimes all you need is a first sip.

[5:13] Get out of your head. Get into your moment. Feel what moves you. Your moment, your flow. Please feel free to write in the comments section which video you liked better, and then we will take a closer look at the details together as a group. In both cases, Cedance 2.5 was selected as the AI video model and that would also be my choice for videos that are 30 seconds long. The advantage of using Hickfield is that it provides you with access to all the latest AI models, and then Chat GPT has to decide for itself which one is the best choice for the task at hand . In my view, Chat GPT made the right

[5:58] choice here, and Luna, which is the more affordable model, generated the video in 720p, whereas Astra created the video in 1080p. This means that the more expensive model in Chat GPT chose a higher resolution. If we examine the structure and the editing of these videos, we can see that this video here consists of seven sequences that were strung together, while the other one is

[6:18] made up of nine individual sequences. And in the video, only music and sound effects could be heard, while in the video, a voice guides you through the storyline and at the end there is also text that is displayed with your moment and your run. The cheaper model in Chat GPT also only generated a very clean one and the expensive one created a complete storyline. Here we can see the two prompts, because that is exactly what Chat GPT is responsible for in this case. Chat GPT writes the prompt and then it selects all of the appropriate settings for you. The prompt from the cheaper model is significantly shorter than the one from

[6:48] the Frontier Model. And I personally have to say, I think this video, which was created with Astra 6, is better than the one with the cheaper Luna model. But we can also learn a lot from it. I am of the opinion that it is not absolutely necessary to burn tokens in CHGB and to utilize Astra for the creation of such simple and basic video projects. You can also easily achieve a better image resolution simply by adding the term 1080p directly into your own prompt. You don't necessarily need the model for that. You can also control the content structure and the overlays yourself using additional information in the prompt. For videos

[7:18] of this type, I recommend that you choose a more affordable model in Chat BeaD, work on the video structure together with that model, and only then pass the final order on to Hixfield for further processing. And I have to say, I find it impressive what kind of results artificial intelligence is now capable of generating for you in one minute and 11 seconds. The video quality that Hickfield has delivered here is already excellent, which is simply due to the fact that Cedance 2.5 has now become an extremely capable video AI model for users. In my view, the most crucial factor for achieving superior results in a video of this

[7:47] kind is not necessarily the AI model within Chat GPT. Rather, you have the ability to provide more context, details, and reference images yourself, which allows you to make a genuine difference in the final outcome. So, if you possess a bit of basic knowledge yourself and develop the storyline in collaboration with the more affordable model, then this will naturally require a greater amount of effort from you, but you will save tokens in Chat GPT and receive exactly the result in Hickfield that you are hoping for. And that is also good news, because we humans are not becoming redundant; instead, the nature of our tasks is

[8:13] simply undergoing a process of change. And then there is one more thing that you most likely noticed while watching this video. The spoken voice struggles with some German words, and this is not because of Shat GPT, but rather due to Cedance 2.5, which is the video AI model being used. This produces significantly better results in English , and for that as well, there exists a very straightforward solution. If you enter a Chat GPT into the very same chat, you will be able to create the video using a voice that is completely accent-free. Chat GPT then connects with Hixfield, and you are provided with a selection of voices that can be

[8:42] used to subsequently rework the video for your project. So here too, it can make sense to combine different AI models. Cedance 2.5 is extremely good for visual representation in the German language, but it is not quite as strong , and you can easily compensate for that shortcoming here by using Hixfield . In the next step, we would like to generate a scene from an AI film together with you. Here I have once again selected Hickfield and then added the following instruction: Create an emotional film scene. A couple says goodbye to each other at the train station. 30 seconds duration, 16: 9. We are now going to generate this once

[9:14] with GPD6 Astra and once with GPD6 Luna , and then we will take a look at the comparison together. The lower-priced, weaker model took 45 seconds, while the more expensive, high-performance model required 1 minute and 21 seconds to complete the task. It's incredible how fast it can be these days. Luna used Cinema Studio Video 4.0 and has generated the video once again in a 720p resolution. Astra, on the other hand, created the film scene in 1080p with Cedan 2.5. And now let's take a look at the result. This is the video from Luna. And this is what the film

[10:16] scene that Astra generated looks like. April rich door. Please write in the comments which one of the film scenes you liked better than the other. From my point of view, the result from Astra is clearly better. There are a few mistakes in this video. The man gets on while the door is still open as the train is already departing, and then there are two men who are seen standing over there at the train stop. And the video was created with Cinema Studio Video 40. With it, you can generate AI films with very detailed specifications

[11:09] in Hixfield. You have a lot of options. For the film setup, you can select the camera, the color, the lighting, and the camera movement, and thus generate videos with a length of up to 30 seconds. This is a very comprehensive tool that gives you many options for creating film scenes in a very targeted manner. And the cheaper model in CHGBT chose this more complex method for video generation, while Astra created the video with Cance 2.5. And if we compare the prompts against each other, the one from Astra is significantly more detailed in nature. Here we see exact timing markers with details about what should occur at which specific

[11:42] point, while this prompt provides more of an overview regarding the scene itself. So Astra has definitely written the better prompt. If you would like to save tokens in Chgd, you should use a more affordable AI model for such videos, but then you must take on the extra work of crafting the prompt in much greater detail together with the AI. And that's not all. We'll also do a third test, and that's where things will get really exciting. Before we get into it, here's my little request again . If you happen to like this video, I would be very happy if you could support me by giving it a thumbs up and

[12:09] subscribing to my channel as well. Don't forget to activate the bell so that you don't miss any more videos on Mondays. And now let us take a look at whether Chb can produce a finished explanatory video together with Hixfiel and if it is truly that simple to implement; that is exactly what we intend to find out right now. Here's my prompt. I chose Hixfield again and added the following prompt: Create a Cärv by showing the most common

[12:30] mistakes made when creating AI videos. The resulting video should be posted on TikTok, with the specific goal of achieving a high number of views on that platform. So, I have not specified a concrete aspect ratio here, but I have given TikTok an indication regarding what it is that I am actually looking to achieve. Then I selected GPD6 Luna again, sent the whole thing,

[12:46] and this is what the result looks like. Your K suddenly looks like a fever dream. It's usually because of these four mistakes. One, you're cramming too many actions into one prompt. Only show one clear movement per clip. Two figures or products change digit by digit. Use a reference image and always describe the objects in the same way. 3. You request zoom in at the same time

[13:05] . Select exactly one camera movement. Four. You expect readable text directly from the generator. Leave areas clear so that you can add the text later in the editing process. The most important trick: first subject, then movement, then camera. Do that before you generate your next clip. So, I'd say this is AI Slope. The video was created with Cand 2.5 and is 30 seconds long. Chat GPT took about 5 minutes to do it. The basic idea isn't bad at first, but the way it is now, the whole thing just isn't really usable in practice. As we can see, producing an explanatory video like this represents a significantly greater

[13:34] challenge than the film scene we were working on previously. And with such a fast model and a simple, short prompt, we were not able to obtain the result that we had originally desired to achieve. But let's see if GPD6 Astra masters the challenge. And this is what the result looks like. Your AI video looks like AI. These four mistakes are costing you better results, and the final one only emerges after the generation process is complete. Too

[13:59] many actions clutter up your prompt. Begin with a clear, visible movement and only add additional details in a targeted manner once the test has been completed. Wine subject moves cleanly but changes its appearance. Please use the same reference image for all further shots and ensure that you describe the figure in a consistent manner. Contradictory camera commands make your clip hectic. Start with a smooth camera movement and test further

[14:26] changes individually for more control. The last mistake happens during editing . You keep every second. Remove the boring lead-in and make sure to start directly with your strongest and most compelling result. The chaotic attempt becomes a clear clip through targeted selection. Please save this checklist and verify your next attempt before you decide to publish it to the platform. There is still some room for improvement here and one could certainly continue editing it further, but I find the result already much better than the first video, and I will now reveal to you what exactly is important in order

[14:57] to achieve the best possible results. But first, let us take a moment to look at what Astra has been up to in this instance. The inexpensive model Luna wrote a prompt which was then used to generate the video in one single pass by means of a separate video model. In other words, the quick solution to the task. Astra first came up with the structure for the video, and then proceeded to have suitable images generated for it. You can also view this directly in Hixfield, because everything that you create with Hixfield via ChatBD will also be saved automatically within your personal Hixfield account. Furthermore, the voice that can be heard in the video

[15:24] was also generated. The individual images that were created beforehand were converted into such video sequences using Minim H3. Everything was then edited together, combined into the final video, and subsequently provided with the necessary subtitles. In this case, Astra proceeded in great detail, step by step, creating the individual puzzle pieces, so to speak, and then assembled them to form the final explanatory video. And that's exactly one of Astra's superpowers. The model divides the task into individual steps and then processes each of these steps one after the other, which is why the price is worthwhile when it comes to dealing with complex tasks. In all three of these examples, we achieved a

[16:00] better result with GPD6 Astra, and the more complex the task was, the greater the difference between the final results became, in my own personal view . However, one must also add that in all three of those cases, we were working with extremely short and rather vague prompts. And now here's my insider tip. If you would like to generate AI videos like the professionals without burning through

[16:18] your tokens, follow this approach. First, develop your video concept in collaboration with a more affordable AI model in ChatGPT. So for all of those back and forth steps where you are not yet quite sure, where you are still developing an idea, or where you are busy creating reference images, just use a cheaper model instead. Then, at the very end, insert everything into a new chat and tell Astra that you would like to generate a video in Hickfield with all of that material. And before that, Astra should optimize everything so that you get the best possible result later. Furthermore, I recommend that if you already possess the

[16:47] knowledge of which resolution you desire, which video, and which AI model is the best, then you should definitely enter that information yourself into your own prompt. The more specific your specifications are, the more work you have to do yourself, but the better CHPD can convert your idea into good stage directions and the more the result will correspond to your expectations. So you develop the idea in significantly greater detail together with a weaker model, then you switch over to Astra to have the result optimized, and finally you bring Hixfield on board to have the entire video generated for you. This way you have a bit more work, but in the end

[17:16] you also receive a better result and are, when all is said and done, more efficient overall. And we can see from this that AI models have become very good indeed, but they still require us humans to intervene in order to steer them effectively in the proper direction for the future. HField can create incredibly good videos. With this, you have direct access to the very best AI models currently available for images and videos. ChatGPT is excellent at managing everything, but ultimately, the decisive difference is made by you, based on what you provide to the AI and the specific direction in which you choose to guide it. And I

[17:45] have another exciting piece of information. For the explanatory video, I took the time to ask Chat GPD how many Hixfield credits both of those videos ended up costing me to produce. The AI Slop video ended up costing me 360 Hixfield credits, and the better version that I produced cost me another 169 Hixfield credits. In this list, we can also see exactly how many individual generations Chat GPT has actually triggered during its time being used in Hickfield. And the better result with more generations was cheaper in Hixfield in the end. If you would like to try out Hixfield for yourself, you can find the link to do

[18:14] so down below in the video description. With this tool, you can create everything from AI-generated images and video sequences all the way to an AI influencer. And if you have any additional requests regarding what I should demonstrate that one can do with Hixfield, or if you have any questions at all, please feel free to write those down in the comments below, and perhaps I will then make a separate video about that topic in the future. Many thanks to Hixfiel for supporting my work here and letting me make my videos the way I want. Thank you so very much for the fact that you have watched this video

[18:38] again until the very end. Have fun trying it out. I will link a Hixfield tutorial here for you, as well as my playlist which contains a collection of additional Hixfield instructions. If you like, feel free to continue watching. And now I wish you a wonderful week and I am already looking forward to when we will see each other here again next Monday. Take care, see

[18:51] you then. Ciao. Ciao.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt