Poor girl doesn't have any idea why half of the tavern stares at her:c
A/N: A V4-preview test. Despite being a test, I still fixed it up, because, well, IT'S FERN! Do you have any bloody idea how long I've wanted to do her ever since I've seen that iconic(not really) scene of her sitting down on a chair? My life, my very essence - all of it was dedicated to this moment. And now? Now, I'm complete.
Now, for my experiences with V4 so far. 1. Character positioning doesn't work as smoothly as I'd hope it to be, but I can chalk it up to a new tech and it'll get improved later on with V5 and future models. It's a nifty feature though, don't get me wrong, and when it works - it works. 2. Due to it being a preview model, most of the crucial parts of the toolset are missing, including advanced samplers SMEA/DYN which make images look "good", inpainting and some others. To compensate, instead, it currently uses V3's inpainting, which means if you need to fix up the characters that got added into the dataset - same Fern, Stark, and just, you know, new characters - you, obviously, won't be able to. Also, same ol' seams, WHICH I REALLY HOPE V4'S INPAINTING WON'T HAVE. As for the lack of advanced samplers - it's not *quite* as bad as it is in V3 where without them the outputs generally look like low-quality garbage, but if you want to use certain styles - yeah, you definitely gonna feel it. 3. Since it's an SFW preview model, it won't do naughty stuff. You can still get something risque like a buttshot here, or an exposed cleavage there, but overall it's neutered to hell. Why? Because the full model's still cooking, and if that means that the depravity I'll be able to get out of it will be on an absolutely different level from V3 - I really don't mind, it's definitely worth waiting and looking forward to. 4. Overall coherency of certain characters that I had huge trouble with in the past(like Nikke's Rapi, V3 was just HORRIBLE at making her) - increased significantly. They look like them and they look great. I cri;0; 5. Natural language understanding is... well, it's hard to say for now, tbh, I need to do way more testing before I can conclude how good it actually is, but for now I'm not particularly impressed. Now, the tag understanding though - oh yeah, it's great. 6. Can generate bottom-heavy characters without spaghetti prompting, editing and inpainting. I think nat lang might be helping quite a bit there. 7. New styles(oh boy, that's gonna be juicy), new characters(NAGA, CINDERELLA, BAY, GODDAMMIT, THE WHOLE NIKKE GALLORE IS HERE! Almost.). And speaking of characters - it seems that even with significantly lower dataset than needed in V3 you can still recreate them, which is great. But again, more testing needed, I'm not fully sure. 8. Despite all the good... Well, let's just say there is some bad, too. Running out of space, but what I CAN say is that without inpainting, it's hard to get anything worthwhile. 9. Oh, and close-ups are of amazing quality:d
FEEEERN
Poor girl doesn't have any idea why half of the tavern stares at her:c
A/N: A V4-preview test. Despite being a test, I still fixed it up, because, well, IT'S FERN!
Do you have any bloody idea how long I've wanted to do her ever since I've seen that iconic(not really) scene of her sitting down on a chair? My life, my very essence - all of it was dedicated to this moment. And now? Now, I'm complete.
Now, for my experiences with V4 so far.
1. Character positioning doesn't work as smoothly as I'd hope it to be, but I can chalk it up to a new tech and it'll get improved later on with V5 and future models. It's a nifty feature though, don't get me wrong, and when it works - it works.
2. Due to it being a preview model, most of the crucial parts of the toolset are missing, including advanced samplers SMEA/DYN which make images look "good", inpainting and some others. To compensate, instead, it currently uses V3's inpainting, which means if you need to fix up the characters that got added into the dataset - same Fern, Stark, and just, you know, new characters - you, obviously, won't be able to. Also, same ol' seams, WHICH I REALLY HOPE V4'S INPAINTING WON'T HAVE. As for the lack of advanced samplers - it's not *quite* as bad as it is in V3 where without them the outputs generally look like low-quality garbage, but if you want to use certain styles - yeah, you definitely gonna feel it.
3. Since it's an SFW preview model, it won't do naughty stuff. You can still get something risque like a buttshot here, or an exposed cleavage there, but overall it's neutered to hell. Why? Because the full model's still cooking, and if that means that the depravity I'll be able to get out of it will be on an absolutely different level from V3 - I really don't mind, it's definitely worth waiting and looking forward to.
4. Overall coherency of certain characters that I had huge trouble with in the past(like Nikke's Rapi, V3 was just HORRIBLE at making her) - increased significantly. They look like them and they look great. I cri;0;
5. Natural language understanding is... well, it's hard to say for now, tbh, I need to do way more testing before I can conclude how good it actually is, but for now I'm not particularly impressed. Now, the tag understanding though - oh yeah, it's great.
6. Can generate bottom-heavy characters without spaghetti prompting, editing and inpainting. I think nat lang might be helping quite a bit there.
7. New styles(oh boy, that's gonna be juicy), new characters(NAGA, CINDERELLA, BAY, GODDAMMIT, THE WHOLE NIKKE GALLORE IS HERE! Almost.). And speaking of characters - it seems that even with significantly lower dataset than needed in V3 you can still recreate them, which is great. But again, more testing needed, I'm not fully sure.
8. Despite all the good... Well, let's just say there is some bad, too. Running out of space, but what I CAN say is that without inpainting, it's hard to get anything worthwhile.
9. Oh, and close-ups are of amazing quality:d