In this post, I want to show you a second option based on DALL-E-3, which is also built-in with chat-based AI interfaces like ChatGPT or Microsoft Copilot. Both of these options require an account, with ChatGPT Plus being a subscription (= not free), but you get so much more than just a bit of image generation. With Microsoft’s model, I don’t exactly know how they deal with their licenses. I have a family account for Office365, which gives you also 1TB of storage on OneDrive, and full access to the current versions of Word, PowerPoint, and Excel. Maybe it’s for free, or maybe it is part of that. There is also a paid version called Microsoft Copilot Pro, which is an add-on to the Office365 subscription, and then the AI is included in all of the office apps — but that is a different use case.
For this post, I will start using Microsoft Copilot because I hope that some form of it is still free.
What is different compared to the Bing image creator? With the chat interface, you don’t write any prompts, but the chatbot does that for you. You will interact with the bot, and tell it what it should change, and it will create a new prompt in the background for you, and generate the images. This makes image generation more like a case where you are talking to a designer, and they come back with options.
Here’s how to get started:
I think you get the point. Please note the part of the bot response where it says “1 of 30”. This is important because you can only have 30 messages per “story” before you need to start a new “story”. What does that mean? It’s a chat, and the chatbot remembers what you told it before. This is important in many use cases, and it increases the quality of your interaction with the bot immensely.
You can tell it what you want to have changed, and it will try to fulfill your request:
As you can see, we can do pretty fancy things, and quite specific things as well. But you also see that it’s not really the “last image” that was modified, but new images being created. Is it possible to modify the existing image instead of creating new images?
Theoretically, there should be a way. When we use random number generation in computing, the numbers aren’t really random. If you fix the so-called seed number, you can generate random, but always the same numbers. And every AI generated image can have a fixed seed, or at least the AI model has that option, but not every interface exposes it.
Let’s ask Copilot:
What an arrogant bot! We could try to convince it, or change the prompts, but maybe Microsoft indeed doesn’t want you to know. Let’s switch to ChatGPT for a change.
I haven’t show the full chat history here because it’s just the same again. A basic request for a picture of a tango couple. And then I ask it for the prompt and the seed. Let’s see if we can regenerate the same image again:
So, unfortunately the ease of the chat interface is not coupled with direct power over the DALL-E parameters.
Let’s see if we can do direct manipulations on the previous picture:
The left picture is the original one, and the right one is the “black and white” version. Not quite, but similar. Looks like the chat interface produced a specific prompt that maintains a bit of a logic, but I think it is important to understand that we don’t do image manipulation, but prompt manipulation only. The text prompt is translated to an image, but the image is not manipulated.
There are so many more things we can do, but I only want to show you one more trick: Let’s create an image of a milonguero dance couple in a social dance scenario, maybe a milonga in Bs.As. I just used that sentence, and this is the output:
The prompt used for generating the images of the milonguero dance couple was:
“A milonguero dance couple in the midst of a social dance at a milonga in Buenos Aires. The setting is vibrant and authentic, filled with dancers. The couple is in close embrace, capturing the intimate and intricate style of milonguero dancing. The man is dressed in traditional attire, and the woman is in a stylish dress, both moving gracefully on the dance floor. The background shows other dancers and the lively atmosphere of the milonga, with vintage decorations and warm lighting.”
And here is my trick. We can get a much more rich prompt from chatGPT by letting it google for it first (left), and then letting that information become part of the prompt (right).
The Milonguero style of tango is characterized by the following features:
Embrace: Close and intimate, often chest-to-chest. The woman’s left hand goes past her partner’s right shoulder, sometimes around his neck, ending up near his left shoulder. The embrace is tight with no space between partners, but without holding tension.
Posture: The woman leans against her partner at a more pronounced angle than his. The man leans slightly toward her but may appear upright.
Movement: Small, rhythmic adornments with the feet, changing weight at the last possible moment.
History: Originated in Buenos Aires during the Golden Era of tango (1935 to 1952). Developed within crowded milongas (social dance venues).
Style: Compact movements, navigable and safe for nearby dancers. Focuses on walking and accessible to everyone.
Expression: Dancers focus on their partners rather than the audience, preferring improvisation over choreography. Emphasizes a feeling of oneness and shared self-expression.
Attitude: Milonguero dancers often adhere to tango etiquette (códigos) for an optimal dance experience.
Distinction: Different from Salon Tango, which involves larger steps and figures unsuitable for crowded social dancing scenarios.
Music and Rhythm: Critical elements include the importance of tango music, rhythm, and the close connection between partners.
Evolution: Some practitioners blend Milonguero elements with a nuevo outlook, creating a style suitable for crowded dance floors yet full of playful interaction.
Based on the characteristics of the Milonguero style of tango, here’s an improved prompt for generating images:
“Depict a Milonguero style tango dance couple at a bustling milonga in Buenos Aires. The couple is closely embraced, with the woman’s left hand past her partner’s right shoulder, possibly around his neck, enhancing their intimate connection. They exhibit a posture where the woman leans into her partner more pronouncedly than he leans into her, creating a dynamic yet balanced stance. Their movements are compact and rhythmic, suited for the crowded dance floor, with subtle foot adornments. The man and woman, dressed in traditional attire, embody the spirit of social tango with a focus on their partnership rather than onlooker attention, reflecting the authentic Milonguero style rooted in walking and improvisation. The surrounding ambiance is lively, filled with dancers who respect the códigos of tango etiquette, making the scene a true representation of Buenos Aires’s vibrant tango culture.”
This prompt emphasizes the intimate embrace, posture, movement style, and social context that are distinctive to Milonguero style tango.
I don’t know if that was much better, was it? It’s still cliche, but the hands are right in both images. And neither looks like what we would expect at a milonguero meeting in Europe… hmmm.
Based on the information gathered, here’s a comprehensive description of Encuentros in tango, which can be used to create a new image:
Definition: Encuentros, also known as Encuentros Milongueros, are tango events that focus on the Milonguero style of dancing. They typically happen in the form of weekend retreats.
Audience: These events are attended by committed milongueros and milongueras who appreciate traditional Argentine tango dancing and etiquette.
Environment: The setting is designed to resemble traditional Buenos Aires milongas, fostering an atmosphere conducive to the Milonguero style.
Dance Style: Dancers engage in close embrace with movements that are compact, musical, and safe for both the couple and the surrounding dancers.
Music: The music played is classical tango, primarily from the Golden Age era of Argentine Tango.
Etiquette and Structure: Encuentros emphasize tango etiquettes like the mirada and cabeceo for dance invitations, and dancers maintain their line of dance (ronda) without zigzagging across the floor. Instruction and correction are not part of the milonga setting.
Seating Arrangement: Leads and followers are usually seated on opposite sides, facilitating dance invitations from a comfortable distance.
Role Balance: These events are typically role-balanced to ensure a good dancing experience for everyone.
Social Aspect: Encuentros often include community meals to strengthen bonds among dancers and encourage social interaction.
Based on this information, I will now create a new image.
The image depicting an Encuentro Milonguero event, capturing the traditional Argentine tango atmosphere, has been created.
I think that is better, and closer to what I wanted to see. What do you think? The key here is to vary the approach to the picture, and iterate a lot, until you find what you had dreamt of in your imagination.
Leave a Reply
Only people in my network can comment.