A little time and some credit purchases are required to get started with the AI video generator. The result is promising.
There’s not just Seedance 2 in life: other models are also excellent for the production of video content, particularly advertising content. This is the case of Kling, which stands out for its ability to maintain visual coherence on several levels. The tool is based on an advanced deep learning technique: a 3D spatio-temporal attention mechanism and a diffusion-type transform model. Using a simple prompt, it extracts representations which serve as a starting point for the 3D spatio-temporal attention mechanism. At the spatial level, the model ensures that each image is visually correct. At the temporal level, it looks for smooth and logical transitions between images. The model also integrates 3D reconstruction techniques of the human face and body. This can lead to interesting camera movements for short videos on social media. Meanwhile, the recently released Kling O is a unified multimodal AI video model. It combines generation, iteration and editing of the image or video. This should simplify the workflow.
We wanted to test this by creating a short advertisement for a fictional brand of tennis balls, called “Dupont”.
6 euros per month and credits to buy
Visit Kling AI. Since December 31, it has been possible to use Video O1 and Video 2.6, with voice control. This plan allows you to generate high definition videos in 1080p.
Registering with Kling AI offers some credits. The standard plan, at around 6 euros per month, entitles you to 660 credits per month. This allows you to generate around ten videos on average, depending on the options chosen. Other plans range from $26 to $127 per month. Based on our experience and customer feedback, the credit system is complex. It is possible to purchase additional ones separately. Given certain problems encountered in video generations, this can quickly become relatively expensive, especially for beginners. To give an idea, Runway, which offers advanced creative control and precise camera control, has a base price of around $15 per month. OpenAI Sora, integrated with ChatGPT, charges at $20 per month. Creatify AI, which generates AI avatars for advertisements, costs around 39 euros per month.
Omni gregi dux is
Once on the interface, click on “Omni does it all”. This section brings together the image and video generator in one place, which is convenient. In order to create our approximately 10 second clip, our strategy is to create 3 separate clips that we will edit later. This should provide greater creative control over each scene. To do this, we will generate images and videos corresponding to the different plans.
The scenario is deliberately simple, given our mastery of the tool. In the foreground, a tennis ball bounces on a tennis court. Once at the top of the bounce, it takes up most of the screen, with a black background behind it. At the end of this clip appears the slogan of our fictitious brand “Dominate the rebound”. This ad is supposed to appear on certain social networks. Note that we only barely integrated the tennis player into this, because various anomalies of movement, or even synchronization with the tennis ball, multiplied during our tests.
First step, we will first create the image of the tennis player, with this prompt, in English, French not being supported by Kling:
"Professional male tennis player, athletic build, 25 years old, short dark hair, intense focused expression, white Nike tennis outfit with yellow accents, holding a professional tennis racket, standing on red clay court, golden hour lighting, cinematic sports photography, full body shot, 8K quality, ultra-realistic"
Among the 4 choices offered, we choose this one. The prompt was generally respected.
We then create the image of the “Dupont” tennis ball. The prompt is:
"Professional tennis ball close-up, bright yellow-green color, visible felt texture and brand logo "Dupont" printed clearly, product photography, studio lighting, white background, ultra-detailed macro shot, 8K quality, commercial product shot"
Among the images proposed, we choose the only one where the word “Dupont” is correctly written.
Note in fact that the Kling text generator is unreliable. We will also use CapCut to finalize the editing and integrate the textual element.
We then generate a third visual, with the clay tennis court:
"Professional red clay tennis court, service box clearly marked with white lines, net in background, golden hour lighting, warm terracotta tones, cinematic sports photography, wide angle shot, 8K quality"
The image generally integrates the requirements of our prompt well.
Expressive videos
To generate the videos, head to the Kling generator. This, thanks to a variable resolution training strategy and a certainly substantial database, makes it possible to generate videos in a large number of formats: 9/16, 1/1 and 16/9. This first format being, for example, ideal for TikTok. It is also possible to add sound, which we will do with CapCut in our case.
The first video should show a “Dupont” ball bouncing on a tennis court. All in a very cinematic way and with a quality image. To do this, we upload the ball image and write a prompt. It is possible to avoid generating certain elements with the word “avoid”:
"@Image1@Image Cinematic slow-motion shot of a Dupont tennis ball striking the red clay court surface in the service box. The ball hits the ground with dramatic impact, creating an explosion of golden clay dust particles that fly upward in all directions. The dust catches the golden hour sunlight, creating a spectacular visual effect. The ball then bounces high into the air with natural physics. The white service box lines are clearly visible on the terracotta clay court. The ball hit is in the court. We can see the tennisman behind the net who has just hit the ball. Professional sports cinematography, ultra-slow motion (240fps feel), dramatic lighting, warm color grading, 1080p quality. Camera movement: Low-angle tracking shot following the ball's descent, then tilting up to follow the bounce. Style: Epic sports commercial, Nike/Adidas aesthetic, cinematic and inspiring, TV advertisement quality. Avoid : two tennis balls, blurred image, low ball bounce, low resolution."
The video generally complies with the instructions.
We then retrieve the image of the ball in the air from the video and request a transition with the image of the ball alone. It is possible to iterate at this stage. The prompt is simple:
"@Image3@Image2 Transition from image2 to image3 in a cinematic way, for a publicity. At the end, the ball must be centered on a black background. The word “Dupont” is clearly visible on it."
The generated video correctly includes the recommendations
Pay attention to the information and images provided
We go directly to CapCut, which offers a free one-week trial offer, to edit the videos and add the text at the end and a sound. After a few tries, here is the final result:
If the rendering is not perfect, especially with the trajectory of the ball and the copyright at the bottom right, it is still interesting and promising. A certain coherence takes place between the different plans. Note that for beginners, the various tests may incur a certain cost. Note also that in terms of security, Kling AI recognizes that it cannot guarantee the security of the data and disclaims all liability in the event of a violation. Downloading sensitive data or identifiable faces should be avoided.
