Same here, and I also hope communities will grow as more and more people move to the Fediverse. Basically it’s the niche parts of reddit that I’m missing but I’ve managed to stay away from it for the most part. I think we need to make sure to generate some content in the communities we are interested, being lurkers won’t help Lemmy grow.
These are great! Thanks for sharing!
I had been using the bot in discord a lot for a while. The prompting is quite different to what it was in 1.5 and 2.1, you should check the Pantheon channel in discord for a good idea of the prompts that gave good results. Basically my experience is that it works very well with simple prompts, and if you want you can also get very descriptive and will get good results. Negatives are almost not needed in the bot, but I think this may be because they were injecting them already, so the 0.9 that was leaked may need them.
Hey, thanks for sharing your impressions.
I haven’t tested it yet. Are you using it with comfy UI?
From what I understand both models could be trained so the current bias may not be a big issue once people start customising them.
Have you used the bot on discord? If yes, are you finding any big differences between this and the bot? How’s it at following prompts?
What is a kudo? And why is your link about dead fish? Was that what you wanted to link?
Edit: OK I just re-read your comment and I guess the Kudos are something for Stable Horde.
What kind of emoji are you after?
Not 100% sure, but from what I read it’s not too far from what 2.1 needs. They have even fine tuned it on normal GPUs. I’m not sure if they will have different versions like they had with 2.1 where they released a 512 and a 768 trained version, which would require less VRAM.
Yes, it was:
Photo of a lemming with a welcome sign, “welcome” written with marker.
No negatives, no fancy adjectives…
I made 8 images and this one was the best (some had no text, some had uglier lemmings, some had misspelled signs) but most were acceptable just not what I wanted. Also the cfg and steps and sampler are random so once we can control that it will be much easier.
Exactly, it’s definitely better than previous SD versions, but not at Deep Floyd level. It can do short words quite well. Some of the beta versions they were trying were better at text, and some were writing too much text (parts of the prompt would become words in the image). Who knows what we’ll end up with, but will be better than SD. This was the best of eight images I made, some had no text, some had ugly lemmings, some had the text with misspellings. I’m sure once we have this in auto or comfy it will be a game changer.
very cool! would be cool if you could add variation to the areas (like a market, port, forts, different architecture, etc) I guess it could be done by combining different images from different prompts.
Hey thanks for this! I’ve been looking for some comparisons with SDXL, this is perfect. I’ve been using mostly DPM++ SDE Karras with low steps and finding it always the best, but it’s interesting that you found that at high steps it’s still high, but not a clear winner. Thanks for sharing this!