Inspiration
Ikebana is the Japanese art of flower arrangement with carefully selected ingredients to convey a specific feeling or emotion.
What it does
User enters: 1 - city 2- month 3- occasion.
The app generates:
1- a text description of a flower arrangement recipe with flowers available in a city in a given month to express an occasion.
2- two images (arrangement 1 and 2), which user may use as 'alternative' visual recipes with the same ingredients, which were generated by the text recipe description.
How we built it
Here are the widgets I used:
3 user input widgets: 1 - city 2- month 3- occasion
Text generation widget for recipe: Returns ingredients and recipe for the flower arrangement.
2 image widgets (arrangement 1 and 2) for two alternative flower arrangements depicted visually with ingredients the app decided for in text recipe generating widget.
Another image widget called arrangement 0 is used as a steppingstone for arrangement 1. This indirection creates visuals that are closer to the desired style and ingredients.
A text generation widget which generates the prompt that is used by image generator called arrangement 0.
A text generation widget, which is used by image generator called arrangement 2.
Static text widget: Describing Ikebana style.
Challenges we ran into
Stable Diffuse XL seems to have limited capacity to translate text to image in detail. For example, number of stems of flowers. So, I couldn't make it to draw what the flower arrangement recipe described. Therefore, I created 2 images as alternative visual recipes. rather than a picture of what the text recipe prescribes.
Accomplishments that we're proud of
Use of 'variables' that are defined in one widget to be accessed by other widgets (e.g., see the prompt of widget called arrangement 1).
Using text generation to create the prompt to be used by the for the image generation (see the widget called arrangement 0).
Use of indirection to generate a more accurate (to ingredient and style) flower arrangement (see the prompt of widget called arrangement 2)
What we learned
Stable Diffusiom XL is limited to translate text to image more deterministically.
Built With
- claude
- partyrock
- stablediffusionxl


Log in or sign up for Devpost to join the conversation.