0 оценок, среднее: 0,00 из 50 оценок, среднее: 0,00 из 50 оценок, среднее: 0,00 из 50 оценок, среднее: 0,00 из 50 оценок, среднее: 0,00 из 5 0,00
Время чтения - 6 мин.

Top-3 neural networks for ad creatives

Today one doesn’t need large creative or design teams since we have neural networks able to generate original ad creatives. So it’s enough to have one expert who can professionally select and enhance ad creatives prepared by artificial intelligence (AI).

We’ve come up with a list of popular services which will help you automate the routine workflow of ad creative preparation. We’ll also share some pieces of advice on how to get a well-composed query for a neural network.

🤖 Midjourney

This one is really popular. Its neural network is very good for details and futuristic stories.

To generate an ad creative one needs to:

✔️ Register on the server in Discord

✔️Open any «Newbies» chat and type «/», then select «imagine»

✔️Enter a query in English after «prompt» — Midjourney will use it to generate an image.

✔️Send the text to the bot and wait, the neural network will be creating 4 images right in front of you

One can create up to 25 sets of 4 images each for free. Then a subscription is required. Below you can see the results for the following queries: «woman wants to lose weight», «woman trying to lose weight», and «woman on diet». Then we add background and some text to get a ready-to-use ad creative.

                                                          Ad creatives generated by Midjourney

🤖 Dream

Generates images based on text queries, the result tends to be quite abstract, but it is generated very fast. The service offers more than 50 styles. Only queries in the English language are supported.

To generate an ad creative one needs to:

✔️Enter a brief description in English

✔️Select one of the available styles and click «Create».

As for the format, it is always a vertical rectangle image. Below you can find the results for the following query: «woman trying to lose weight» 👇

                                    Ad creatives generated by Dream

🤖 Craiyon

This neural network visualizes any idea, but the images contain broken pixels and blurred details, thus they cannot be considered ready-to-use ad creatives. We suggest users either try to improve the text query or forward the images to designers who will enhance them. Below you can find the output images for the following query: «woman was able to lose weight».

To generate an ad creative one needs to:

✔️Open their website

✔️Enter a text query in English

✔️Get a collage of nine images

Ad creatives generated by Craiyon

How to formulate a correct query for a neural network to get an ad creative 

  • 1.Object

It is the core part of every query which will be placed in the center of the ad creative^ this is why it needs to be well thought out. For instance, a «woman», «doctor», «money», etc.

When you need to put a few objects on your ad creative, then provide the exact number, e.g. «three women». If you use a plural form without specifying the exact number, then the neural network will generate a random number of objects. Keep in mind that for a neural network it is hard to process big numbers of objects (more than 5 or 6).

  • 2. Negative forms

Avoid such words as «no», «without», «excluding», «except for» and similar. Neural networks are very straightforward and ignore any negative forms, e.g. for a «man without a beard» text query you’ll get an image of a man with a beard. So you should better type «clean-shaven man» to get a relevant image.

  • 3. Actions

Formulate an action in such a way that it can be easily put in an image, i.e. a verb «sits» is much better than «dreams of» or «matters». When you still need a verb which is not an easy-to-depict action, then try to add more details, e.g. instead of «thinks of» type «sits immersed in thoughts».

  • 4. Description & details

It is important to find a balance here. One may think that the longer and more detailed the query is, the better the results are. While the developers warn us that excessive details tend to overload the system and the output images will be far from the original idea. Though you should add adjectives to the description, they’ll significantly enhance the image, i.e. a «happy / sad woman», etc. To select relevant adjectives in English you should check Lose the Very website.

  • 5. Specific details

Clearly formulate what you expect to see on the image, i.e. if you want to get an image with a woman by the seashore who is standing on the scales, then be specific. For instance, instead of  «woman on beach stands on scales», try out  «sad woman on beach standing on scales».
Besides, you can send a reference image to the network along with its description. And be careful, Midjourney has a blacklist of words which cannot be used to generate images, e.g. anything linked to nudes or debauchery, so you should consult The Midjourney Banned Word List beforehand in order not to be banned.

  • 6. Use special services and other neural networks 

To save time on coming up with the right query, you can use a chatbot with AI,  ChatGPT, and ask it to formulate a query based on provided parameters, then copy the output description and paste it into the neural network to check the results.

Another service able to help you select a description for your query is Lexica, a website for image search where one can find a relevant image and copy its description.

Despite the fact that today’s neural networks are still imperfect, they are already able to speed up the workflow and provide us with ideas thus optimizing the main resource we have — our time ⏳

Оцените статью
Написать комментарий