Improve sticker generation smart picker

[DaphneMuller]

Describe the feature you'd like to request

@lukasdotcom created a smart picker to generate stickers using image generation. It currently looks like this and I think the design would benefit from improving this in a next iteration:

The user select a "sticker generation" smart picker, then it pre-fills the prompt section to "cartoon, neutral background, sticker of" and then the user fills in the rest
This string is by the way translated into the language of the user.

Describe the solution you'd like

In the Image generation UI, create a style selector. Depending on the selected style, inject the prompt ("cartoon, neutral background, sticker of" and hide this prompt from the user.
@marcoambrosini please fill this in according to your wishes

[DaphneMuller]

Assigned @marcoambrosini for review

[marcoambrosini]

Hi Daphne, from a design standpoint, I would remove the image from the left navigation. It does not fit well that component.

I also would not create a dedicated UI for sticker generation, but instead integrate it into the image generation tab.

[DaphneMuller]

thanks for your feedback. Could you clarify what you mean exactly with "I also would not create a dedicated UI for sticker generation, but instead integrate it into the image generation tab.", do you mean to keep the UI as is for sticker generation except from removing those images from the left navigation bar?

[DaphneMuller]

I just had a call with Marco and we settled on not creating a new UI for sticker generation.

Instead the suggestion is to add a style selector in the current Image generation design, just to hide the prompt for the sticker generation in there, and we leave everything else the same.

@marcoambrosini will provide a mock up tomorrow
@lukasdotcom you can start on that after Marco provided his input

[julien-nc]

add a style selector in the current Image generation design, just to hide the prompt for the sticker generation in there

Remember the form is generated from the description of the task type. This description for TextToImage does not include a "style" input.

• We can't change the task type (this could break previous tasks).
• We could add an optional parameter in some TextToImage providers but then we would not be sure the feature is available (to decide if we register the sticker gen file picker)

What we can do is add a new "TextToSticker" task type. As this is just a "custom prompt feature", I think we don't need to implement multiple providers (in each AI apps). We could just implement a generic provider for this task type in the assistant. This provider would just schedule a TextToImage sub task with the modified user prompt (a bit like the AudioChat Assistant provider just schedules 3 sub tasks).

Wdyt @lukasdotcom ?

[lukasdotcom]

add a style selector in the current Image generation design, just to hide the prompt for the sticker generation in there

Remember the form is generated from the description of the task type. This description for TextToImage does not include a "style" input.

* We can't change the task type (this could break previous tasks).

* We could add an optional parameter in some TextToImage providers but then we would not be sure the feature is available (to decide if we register the sticker gen file picker)

What we can do is add a new "TextToSticker" task type. As this is just a "custom prompt feature", I think we don't need to implement multiple providers (in each AI apps). We could just implement a generic provider for this task type in the assistant. This provider would just schedule a TextToImage sub task with the modified user prompt.

Wdyt @lukasdotcom ?

That probably would be simplest.

[marcoambrosini]

Hi, after trying some alternatives I think the best course is to do what @DaphneMuller suggested before. I can't come up with something simple enough that improves that. So, to reiterate it would look like this:

• Create new "generate sticker" action under "work with images" group
• In the confirmation button, instead of "submit", replace the string with "Generate sticker". While we're at it, in the regular image generation we can replace "submit" with "generate image".
• Remove the prompt from the input field