perplexity

Hi there @[email protected],

Saw one of your posts in here and thought I'd give you a little cheer.

For what it's worth, you're making a big difference with Perchance, and you should not beat yourself up for lagging behind in terms of latest technologies. Perchance for me personally was the first place where I've delved into the world of AI generation that didn't limit me in terms of creativity. Thanks to you I've gotten to know a bit more about Stable Diffusion and prompting in general. This was a kickstart to explore it further. Without perchance I would have probably just shrugged the whole AI generation thing off and called it a day.

Thanks to you and perchance, I've found a community that I talk to this day. It's been more than a year now. And this is, my man, something for you to be proud of. You created something that connects people. The comments plugin comes with its own challenges and creepy people, but this is an absolute killer feature that distinguishes perchance from other AI platforms. Real time communication while generating is a unique and amazing way to connect with others. Forum posts just don't cut it today, because everyone is used to instant messaging these days.

Don't listen to people who complain about perchance being horribly out of date. These people don't know what they're talking about. They didn't lay a finger to improve perchance in any way. Pulling off a project like perchance is not an easy feat. Not to mention that you do this voluntarily, and it costs nothing for us. Take all the time you need, and take pride in your work - because it does makes a difference.

Hugs from furry generator, Perp

Hi @[email protected],

I was wondering if you're planning to make it possible to access elements inside iframes (such as userInputs, comments, gallery)? Using an editor can be pretty limiting at times. It would unlock so much new ways of customizing your own generators if one had access to elements inside iframes to work with them using JavaScript/HTML/CSS. I was just working with some of the user inputs, and didn't realize that even these are inside an iframe as well... which is quite a bummer. I understand that it probably makes sense from architectural standpoint, though.

Thanks!

Hi,

I would like to propose an interesting UX change for text to image plugin. I want to propose to shift from "public by default" approach, because it leads to ease of misuse of it. Generator owners have to deal with people who do not respect the rules of the generator and pollute public gallery with images that are either repetitive or just inappropriate

When you're new to image generation and perchance, it's quite easy to get lost. Public gallery often ends up being a playground/scratchpad as opposed to showcase of the user's best effort. A lot of people do not bother switching to other galleries. Some people also just do not read the chat (or do not participate), but still upload images and end up getting banned without even knowing that the moderator gave them a verbal warning in a chat. It can create tension and misunderstanding between the staff and the visitor.

What I want to propose is this:

• Upon first visit to generator, Perchance could ask the name of the gallery which would act as a "private" gallery for the user.
• Such gallery becomes the default gallery for the user on this generator (meaning that the name of the gallery will be pre-entered in the saving image modal window input field for the name of the gallery, like public currently, as well as it would open by default, again, like the public one currently does).
• The pill for the public gallery should be made impossible to close, and should always come first in the list of quick access galleries.
• The pill for the user-defined "private" gallery could also be made impossible to close (until the local storage is cleared), and should come second in the list of quick access galleries.
• Other galleries for the user could be closable.
• Users should be disallowed to upload into public gallery directly. Using public as name of a gallery to upload image to should be an error. Instead, users have to go through their own galleries and click a button that would duplicate generated image into public gallery, simply by the click of such button. No need to enter public anywhere, just click the button and the image is copied into public gallery immediately. Such copy should be allowed to happen exactly once. The button should also be available to click on/tap right away without having to hover, like upvote/downvote buttons and the others currently are.

I personally think that this approach would allow newcomers to feel safer while experimenting, yet still be able to get feedback on their prompting and encourage people who are already experienced in prompting to be more picky with the images that they want to share with everyone. Both sides are happy.

Hi @[email protected],

I'm here again to give you a few more suggestions for UX improvements and rehash some old ones.

Improve automated NSFW tagging of images. It's not uncommon to see clearly NSFW images in PG13 galleries, which usually end up there by accident. Sometimes it happens by bad actors, too. Very often misspelled tags end up being interpreted correctly and thus bypassing the filter. Some tags can be quite ambiguous too (think balls 😁), you can't just simply forbid it by dictionary lookup. Why not employ AI for this purpose? You already seem to rely on AI alot, and extra request doesn't seem like it would add up alot to the costs. You can cache the GPT response to avoid having to send the request for each generation with same tags.

Implement searching by tags within a gallery This has been a recurring request of how convenient that would be to search by tags inside a gallery. I understand to implement properly you would have to ideally to store everything in the database, which I remember you were reluctant to touch at this point. Are you the person managing T2I framework? In this case maybe you can wind up with the simplest of searches on user side, loading all pages of the gallery and caching the tags on a page (I remember there were saved inside an attribute for each generated image). Then on client-side just do simple regex search and filter images.

Also while I'm at it, I have few questions regarding generator management:

Do you plan on adding (more) hooks for modifying comment box and a gallery? Now that I got to play with generators a bit, I get it now that accessing data from an iframe is painful, if not impossible. I understand the idea for encouraging generator creators to code, but reimplementing chatbox or a gallery just for simple change in CSS is IMO not justified.

Having experienced managing generators with other people, writing a code becomes painful in that one can easily overwrite other's work. Not asking for cloning of git functionality, but there should be at least some sort of visual indication for when someone is working on generator's code to discourage other people from changing code while someone else is working. Plenty of times someone had accidentally canceled someone else's work, which is not very fun.

Thanks for keeping perchance alive.

Hey again,

I just had an idea - given that you've recently implemented creation of rooms for comments plugin (which is quite cool), I think now it would be more or less trivial for you to implement "private" messaging (I say that in quotation marks, because channels are still technically public). You can hardcode the slash command with the syntax /invite <user id> <room name>, which would then trigger the API call as if both the inviter and invitee had created the same room. Alternatively, a smoother approach for this could be not triggering any API calls, but just display a system message of a sort to the invitee that they were invited to a room, and they would have a choice if they want to join the room or not.

Hello @[email protected],

Thanks for implementing posting to multiple galleries, this has been very useful! Looking forward for model updates. I came up with short list of further improvements for the UI.

Gallery:

• Add a button to 'share' already existing generated image to another gallery in the same generator (reupload it to another gallery on the same gen).
• Add the ability for the image owner to 'delete' the image from the gallery (maybe implement it as giving the image lowest score possible). It's quite absurd that the image authors can not delete their own creations, or at least hide them from everyone.

Comments:

• The way how custom room creation is implemented is IMO currently quite silly. Anyone can post arbitrary amount of rooms anywhere, and all hundreds will be added as tabs automatically (even if temporary). Also the attention of the user is not drawn enough to the newly created room. IMO, it makes sense to instead underline the room as if it is a link, inviting a user to click. On click, the room would open, even if it was previously open - then just switch focus to that room.
• Make it possible to share galleries within comments in form of a link, clicking on a link should add a tab on the gallery quick access you've recently implemented and open the gallery.

P.S. t2i by default uses a gigantic emoji pack, which is really great and has wide variety of emojis that are really fun. I assume these were downloaded from one of these sites that offer free emojis/stickers? Any plans to update them? I know that gen owners can add their own, but all popular gens are owned by the staff (or so the story goes). Would you add a small emoji pack I've made into the default list that t2i uses?

weasel_eww = 78b0b8619a440661dc7eae5b6b0b8174.png
weasel_angery1 = 09c65257bce2dbcbff3d4199da1924f3.png
weasel_what1 = 08585158becf45446541832f7830aa04.png
weasel_surprised = 841e86f81d6aede99e25402bfbb3c935.png
weasel_tired = f90a0fdc2831bfeea2b00e13b4c1794f.png
weasel_angery2 = 80a58b1eee31b4b9d9ec5b9e55f92d6a.png
weasel_sleeping = 62da5d73e80cdb67ab2919753eb96926.png
weasel_what2 = 2b08484dc2339ec37a71627c026f6222.png
weasel_happy1 = 60c16d39b27ea3689175eb57108b1c76.png
weasel_curious2 = a784e1ddf86ad3cc715b68c0b8c2e18f.png
weasel_curious1= 41f00b405892f5888b21baa991297b7c.png

Hi @[email protected],

What's your opinion about making it possible to post at multiple galleries at once? Currently, we have to use seeds, which works but can be cumbersome, especially on mobile. It would be nice to have the ability to put multiple gallery names separated by commas when saving without having to think about seeds.

Hi @[email protected],

I have another suggestion. Whenever the new message arrives to the channel that is not a current channel that the user is in, the tab should get highlighted with some color. General channel is the one that the users land in when they open a generator, and they tend to spend most of their time there, ignoring other channels. It often leads to people posting questions to General instead of appropriate channels simply because these other channels are ignored. It can sometimes get disturbing when people start talking about 18+ topics in General instead of appropriate one. General should be kept for general discussion. I think that this feature would be a good step toward guiding users to appropriate channels without other users gnawing at them.

P.S. Report message feature was much needed, thanks for implementing it. A slight nitpick - the shadow/blur in it can be more transparent. In light mode, it does not fit well.

Thank you! :)

Hi there,

Since @[email protected] was OK with bugging them with features, I'll seize the opportunity. Previously, I've published the list with improvements earlier (https://lemmy.world/post/17986442), which was received well, but I see how it can be overwhelming. I wasn't sure if it was a good idea to continue posting in that thread or make new one, so I'm making a new one. Feel free to merge threads if forum staff deems this necessary.

I picked a few low-effort features out of that list which I think could be considered essentials, and which would be great to see pushed upward in your TODO list, namely:

Prompting

• Introduce copy button in prompt and anti prompt fields for better UX on mobile (click to copy). Though, since I do not own any generators, I'm not sure if this is something that generator owner could do on their own?

Gallery

• On mobile, thumbs up and thumbs down buttons on images in the gallery should show up without having to tap on an image to reveal these buttons to make it easier for people to upvote or downvote.
• Implement search/filter in galleries (at least most simplistic, regex-based search by prompt and saveTitle/saveDescription). If you rather not touch database at all at this point, since prompt is stored in HTML after an image is generated, could be a simple JS script that searches within image containers as a starter. Users could manually load gallery pages they wish to search in. Though, is this also something that the owner of the generator could potentially cook up on their own?
• Store the galleries that the user visited earlier in local storage, so it is easier to switch between them when saving or browsing. The galleries could be presented as dropdown listed sorted alphabetically ascending.

Two items from that list are potentially the ones that can be implemented by generator owners, please enlighten me if this is the case - I've seen some very fancy looking generators, but wasn't sure if most of it is just CSS. Thank you, and looking forward - it's always cool to see Perchance evolving!

On behalf of furry-ai gen community I would like to present a list of UX/UI improvements suggestions, which I hope users of other gens will find useful too.

Prompting

• Introduce copy button in prompt and anti prompt fields for better UX on mobile (click to copy)
• Change anti-prompt field to be textarea (the one you can drag to expand)
• Fix bug that makes it impossible to save a landscape-oriented image to the gallery without resorting to changing phone's screen orientation (heart button is out of bounds, and I'm not the only one experiencing this)

Gallery Plugin

• On mobile, thumbs up and thumbs down buttons on images in the gallery should show up without having to tap on an image to reveal these buttons to make it easier for people to upvote or downvote
• Introduce double tap gestures for mobile users on images in gallery to make voting even easier. Double tap to upvote, double tap on upvoted picture to remove the upvote. Downvote could be without a gesture.
• Implement tagging an image with user ID to prove identity of a poster, which may show up as a flair akin to one that the chat box has as an opt-in (by default all posts to the gallery are anonymous). Currently people tag their work with saveTitle and saveDescription, which is easy to impersonate. It would introduce some complexity for blocking though, then it could be treated in the following manner: blocking anonymous post of user A will block all all further anonymous posts of user A from the gallery, blocking non-anonymous post of user A will block further both anonymous and non-anonymous posts of user A from the gallery.
• Introduce public/private galleries, public galleries being an opt-in (that is, by default gallery is private). It would be nice to have public galleries show up in a dropdown list when posting an image to the gallery or when browsing a gallery. At the very least, it would be nice for perchance to store the galleries that the user visited earlier in local storage, so it is easier to switch between them when saving or browsing, again KISS implementation could be just a simple dropdown list.
• If public/private galleries end up being implemented, consider adding a description field to them.
• Add the ability for the user to delete the images that they upload to the gallery. It's quite weird that this functionality does not exist.
• Blocking button 🚫 and hide image button with crossed eye icon are currently too far apart from one another, which makes it difficult to use because they both represent a form of blocking yet they are visually too far away from one another. From my experience people just go straight up blocking the poster and not the image. Perhaps the concept of blocking could be implemented by a click of a blocking button, and pop up being displayed afterwards asking what the user wants to do (block the user or block the image), or at least group the "hide image" and "block poster" together.
• Implement grid view of images on mobile (similar to what Instagram has). On click, the view could change back to the 1-image-per-row view (again, like on Insta)
• Implement search/filter in galleries (at least most simplistic, regex-based search by prompt and saveTitle/saveDescription)

Comments Plugin

• Make all https://generated-images.perchance.org URLs to appear as hyperlinks in chat so that they are clickable, many times it gets annoying having to copy URL manually, even if its within perchance.
• Add the ability to upvote (and possibly downvote) the messages in chatbox
• Add the ability to edit your own messages in chatbox

Other

• Add the ability to talk other people's generated characters found in gallery, not only your own during generation (possibly only with permission of the person who generated the character? Technically prompt is public to everyone, and given that the AI deduces the personality from prompt, it shouldn't be an issue)