Jasmine Rodríguez, Julia Kadie, Sreya Guha, Olivia Otto, Sheikh Srijon
Artist’s Statement
We set out to create a game that makes use of newly developed AI tools while adhering to the fun, lighthearted values of classic games like Charades and Pictionary. ChatT.P.G. (The Prompt Guesser) allows you to exercise your creativity by coming up with outlandish prompts to feed into DALL-E, an AI system that generates images. Players select an adjective, noun, and environment card to guide their prompt. Opposing teams have to guess the original prompt based on the image DALL-E outputs.
ChatT.P.G. captures fellowship and challenge. Players work together with their teammates to come up with their opponent’s original prompt based on DALL-E’s output, and aim to devise prompts that generate challenging images for the opposing team to block them from gaining points. Players only know that a prompt contains an adjective, a noun, and an environment. They are challenged to fill in the gaps between this limited information and the resulting image. On the other hand, this is a game that can be played again and again. Though players draw cards to guide the creation of their prompt, there are an infinite amount of prompts to be created, making it a game that can be played over and over again!
Put on your creative thinking hat, grab your friends and a computer, and have fun!
Image of DALL-E
Concept Map
To begin the design process, we brainstormedthe mechanics of our game. In doing so, we created the following concept map. This concept map helped guide our decisions on our game’s core formal elements and values.
Concept map
Initial Decisions on Formal Elements and Values
Initial Formal Elements
- Players: Initially, we decided to make this a team-based guessing game. We divide players into 2 teams of equal size. Teams work together to guess the prompt that an individual on the other team came up with.
- Objectives: The objective of the game is to gain the most points through having the most accurate guesses. Since each prompt has three categories, teams get one point per category guessed.
- Rules and Procedures: 1 player from Team 1 begins as the “prompt generator”. The prompt generator selects one card from each of the three decks (one adjective, one noun, and one environment card), and then generates a prompt to type into DALL-E based on their cards. The prompt generator should keep their cards hidden from Team 2. For example, if the player selected: “Emotion” “Animal” “Planet” they may come up with the prompt “Happy dog on Mars”. The prompt generator has 1 minute to come up with the prompt. The prompt generator then shows Team 2 the resulting image. Team 2 has 1 minute to come up with a prompt guess. The prompt generator will then reveal the prompt. Team 2 will get 1 point for each word of the prompt that they got correct, up to 3 points. For the next round, the roles switch: the prompt generator comes from Team 2, and Team 1 are the guessers. Prompt generators should rotate each round.
- Resources: Players have our game’s deck of word-type cards and 1 laptop per team for using DALL-E to generate images from prompts.
- Boundaries: The game stops after players end the game. Images are not intended to be too personal, so players can exist independently within the game.
- Outcomes: The outcome of who wins the game is determined by the number of points teams have gotten. The first to 10 wins.
Values
In our game, we focus on facilitating 2 types of fun: Fellowship and Challenge. Our game is mostly about being social and funny with your friends. While the outcome of the game centers around earning the most points, the primary goal is to make your friends laugh. We hope to leverage fellowship aspects to build social ties that last once the game ends.
The second type of fun is challenge. At this point in time, there still is not much skill involved in prompt generation and guessing, so players are winning just for the sake of winning.
Testing and Iteration History
Playtest 1: Tuesday Week 3 in class with other students in CS247G
Playing cards from our playtest
Our first playtest unveiled a few key insights. First, players loved generating funny prompts and smiled a lot while playing our game. Second, this playtest confirmed our initial assumption that the adjective is the most difficult to guess in this game.
Additional insights resulted in changes to our game:
|
Insight |
Change(s) |
|
DALL-E takes a minute or two to generate, which feels like a long time in a game. Waiting around for DALL-E to finish generating each set of images disrupted the flow of the game. |
Have both teams generate prompts at once at the beginning of the round. [Procedures change] |
|
Players were confused about who should write the prompt. Players wrote the prompt as a group instead of individually as originally intended. This presented complications because players were trying to collaborate silently without giving anything away to the opposing team, leading to sub-par prompts. |
Make the distinction between collaborative activities and individual activities clearer in the rules. Guessing is a collaborative team effort, while generating prompts should be an individual effort. [Rules change] |
Additionally, on this iteration of our game, we received feedback from the teaching team about how players will judge closeness between a prompt and a guess. To address this concern, we decided to add language to our rulebook that tells players to be generous with their awarding of points, so that if someone guesses “happy” and the prompt was actually “elated”, they will still award points [Rules change].
Playtest 2: Thursday Week 3 in class with other students in CS247G
Once again, this playtest resulted in a few ideas on how we could improve our game:
|
Insight |
Change(s) |
|
Even with both teams generating images at the beginning of the round, the slowness of DALL-E’s image generation still creates an awkward period of downtime in the middle of the game. |
At the beginning of each game, have each player on each team draw cards and independently generate a prompt and put it into DALL-E on their own devices. That way, with teams of 4, for instance, 8 rounds of play will be ready by the start of the game. With this added computing resource, rounds will move much quicker and players will spend less time waiting around. [Procedures change, Resources change] |
|
Players added lots of detail to the prompt and sometimes placed the environment within the noun phrase, when we had originally intended prompts to be in the order of adjective, noun, environment. In particular, when prompted with Noun: “Politician” Adjective: “Politics” Environment: “Ocean”, one player generated: “Joe Biden giving a speech on the sea levels and so are we”. This prompt was very difficult to guess because it did not follow the expected semantic formula. |
First, require that players put their words in the order “Adjective”, “Noun”, “Environment” with prepositional phrases in between. This will help make the guessing process more straightforward. [Rules change] Second, we changed the design of the playing cards to have a number on it so that information about rules is more distributed. For example, an adjective card has the number “1” on it, reminding players that the adjective they come up with should be at the beginning of their prompt. [Design change] |
At this stage, we also addressed a concern from the teaching team. The teaching team was concerned about the game being too easy. However, after our second playtest we see that while the noun phrase is relatively easy to guess, the environment and adjective are very difficult to guess because players err on the side of creativity as they have incentive to make the prompt not easily guessed. In addition, adding constraints like certain words changes the dynamics of the game, making it too challenging.
Playtest 3: Tuesday April 25 in class
At our final playtest, we gathered the following insights which led to our final game changes.
|
Insight |
Change(s) |
|
We noticed that constraints on how to write a prompt were not clear. Even though players put their words in the right order (adjective, followed by noun, followed by environment) players made their prompts too difficult to guess. This included adding additional adjectives to qualify the environment, or writing an abstract description of a noun in their prompt (e.g. “rounded bouncy object” instead of “ball”). |
Add very descriptive image of format to the back of the box [Resources change, Design change] Add a variety of examples to the rulebook so players better understand what their final prompts should look like in terms of structure and simplicity. Emphasize that a single entity should be chosen for each part of the prompt, and explicitly ask players to refrain from adding unnecessary words or abstractions to their prompt. [Rules change, Design change] |
Prompt “century old transportation device inside a hospital morgue” was too difficult to guess, even according to the player who originally generated the prompt.
Prompt “wear a green dress at Monster’s university” did not follow our expected prompt format.
At this final stage, we addressed final notes from the teaching team. The teaching team wanted us to think about the physical boundaries and “magic circle” in our game. In particular, when the guessing team is working together to guess the prompt, what is the other team doing, especially the folks who are not a part of developing the image in question? Through our playtests, we found that other team members on the non-guessing side were really engaged in listening to the guesses from the guessing team, since the guessing team brainstorms aloud. Since these team members also didn’t know the exact prompt, they were equally invested in the process of figuring out what the prompt was.
Initial Prototypes: Box and Cards
Original Design
From left to right: front of box, back of box, example playing cards
In our initial design, we used a cartoon style robot and a warm color palette of reds, oranges, and yellows. We put a robot painting a picture on the box cover because we felt it demonstrated how our game combined technology and artwork through the use of DALL-E (written on the robot). The cards were kept as simple as possible; we color coded them so the three categories could be easily distinguished. Additionally, we wanted to communicate what a round in our game looked like on the back of the box through an example and instructions.
Second Design
From left to right: front of box, back of box, example playing cards
In our second iteration of our box design, we pivoted quite dramatically. We had received feedback that the initial design seemed too juvenile, and we also received the recommendation to make the design more tech-inspired/futuristic rather than cartoonish and artistic in order to more clearly communicate the theme of the game. In a playtest, we received feedback that it was confusing to remember what order the prompt structure should go in (adjective then noun then environment). In order to address this concern, we added numbers to the cards to reduce the amount of information that players need to keep in their working memory while playing. By distributing more information across the artifacts, we allow the player to rely on recognition rather than recall. We also removed the tagline comparing our game to Pictionary as this didn’t clearly communicate what our game was about.
Final Prototype
Box
From left to right: front of box, back of box
For our final box design, we made a few key changes. We swapped out the DALL-E screenshot of firefighters on the front for graphics that matched the rest of the front cover’s style, and we also changed some of the text to a sans-serif font to increase readability. Another big change that we made was to adjust the wording on the back of the box to use language that sounded more like a sales pitch (and therefore more like a game box cover), since the previous text just detailed out the instructions in a very straightforward way.
Cards
Example cards. Left to right: adjective example card (back, front), noun example card (back, front), environment example card (back, front)
For the final version of the cards, we pivoted to a simple black and white theme to reflect our changes in the box. We added little robot icons with different colors to make it easy to differentiate between the three categories of cards. In addition, we added numbers (corresponding to ordering) to both sides of the card. We wanted this to mirror playing cards and draw on a template that users may already be familiar with. The numbers on both sides allow users to see before and after drawing a card, the order words should go in.
Print and Play
Our final print and play is available here. Cards have a front and a back side, so pages (1, 2), (3, 4), (5, 6), … should be printed on the same page (one on the front side, one on the back side).
Rulebook
We created a finalized rulebook that incorporates changes based on the key insights from our playtests. This would be the “Instructions” included in our game box. Our final rulebook is linked here.
Video of Final Playtest
We recorded our final playtest and it is posted on this unlisted YouTube video: https://youtu.be/0jZLfyWXmx0.
