This image is the final result after some minor edits to the toes on the right foot and the ankle on the left leg then upscaling with tiled diffusion and only includes the upscale parameters. Here are the parameters for the original txt2img generation:
Prompt: kirisame marisa, solo, sitting, on ground, (barefoot:1.2), (soles:1.2), knees up, braid, toe scrunch, (foot focus, foreshortening:1.2), legs, blonde hair, yellow eyes, witch hat, black dress, waist apron, neck ribbon, hair ribbon, messy hair, hair between eyes, glowing eyes, dark background, somber expression, depressed, looking down BREAK dark energy, reflection, colorful, vibrant colors, portal to another world, limited palette, blue Negative prompt: (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, 2girls, (watermark:1.2) Sampler: DPM++ 2M Karras Seed: 1387458770 Steps: 20 + 15 Hires Cfg Scale: 7.5
very detailed, 1girl, solo, full body, (girl stands at headquarters, (points to astronomical map) with her hand, astronomical map on wall), glasses, medium skirt, bob cut red hair, astonished look, dynamic angle, Negative prompt: [(negative_hand-neg:1):0.2], (worst quality:1.4), (low quality:1.4), extra limbs, text, name, signature, logo, bad anatomy, disfigured, watermark, jpeg artifacts, bad eyes, ((lips, tooth)), navel, letterbox Steps: 30, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 2558751036, Size: 1920x1080, Model hash: 2523c9050f, Model: breakdomainxl_V05g, VAE hash: 551eac7037, VAE: sdxl_vae.safetensors, Clip skip: 2, Version: v1.6.0-2-g4afaaf8a, Hashes: {"vae": "551eac7037", "embed:negative_hand-neg": "73b524a2da", "model": "2523c9050f"}
Hires with Kohya Hires.fix in size 1920x1080.
Then processed with DarkSushiMix and corrected the hands by inpaint, and then did FHD->4k (multidiffusion with regional prompt) and corrected the hands by inpaint again.
upscale is a little blurry but these are some high quality pirates nonetheless, would plunder with
Yeah, I'd love to be able to generate higher res, but my poor little 3070 runs out of ram if I go higher than 1280x1280 or 1024x1536. I've tried the SD upscale script, but it tends to throw too much cursed looking noise into the image, so I stick to ESRGAN to smooth over some of the leftover diffusion noise.
Dein Werk gefรคllt mir sehr gut... Ich ill mich auch in AI-Art ausdrรผcken.. Verstehe aber die Negativen Prompts nicht und wรผrde dich gerne fragen wofรผr diese stehen???
Dein Werk gefรคllt mir sehr gut... Ich ill mich auch in AI-Art ausdrรผcken.. Verstehe aber die Negativen Prompts nicht und wรผrde dich gerne fragen wofรผr diese stehen???
Good picture, only glaring issue is where the legs cross and maybe the feet but they're tucked away so it's less of a concern. Everything else is very lick-able.
The name is a reference to a poem by Indonesian poet Sapardi Djoko Damono about the immortality of man and the ephemeral nature of time.
Yang fana adalah waktu. Kita abadi: memungut detik demi detik, merangkainya seperti bunga sampai pada suatu hari kita lupa untuk apa. โTapi, yang fana adalah waktu, bukan?โ tanyamu. Kita abadi.
The skirt is so tiny, she even let people rape her, she can't hide that kind of asset, she should let people touch it and grab it as much as they want, and of course lick it
I prompted it on 2023-10-30 but it took me a while to finish. She's not a halloween cosplayer though, but a real blood moon priestess (meaning she cosplays whole year round).
Yup, it's Lily the cum-summoner (damn, I should have made her eye color consistent with the parent/sibling images). Between this and the cum-angel from post #50167 I'm starting to assemble a whole cum-lore.
@zakiry It was some photorealistic models block merged with darksushi (and maybe some additional anime models? I don't remember). If you want it, I posted it on https://civitai.com/models/11790
"Status: Deleted : it has poor quality" sooo oooook? should i go 8k or 16k quality ?
bye everyone
Resolution is not the only criteria for quality. -highres status:active brings up 19.6k non-deleted posts and counting. If you need feedback on what is wrong with individual uploads, you can ask in topic #70.
@zakiry It was some photorealistic models block merged with darksushi (and maybe some additional anime models? I don't remember). If you want it, I posted it on https://civitai.com/models/11790
@zakiry It was a test version of antlersmix_11. I must have forgot to tag it, fixed now. I didn't end up liking this mix so I didn't post it anywhere, but here most of the style comes from the loras (mainly dreamwave - it's on civitai) so any semi-realistic model should look similar I think.
@zakiry It was a test version of antlersmix_11. I must have forgot to tag it, fixed now. I didn't end up liking this mix so I didn't post it anywhere, but here most of the style comes from the loras (mainly dreamwave - it's on civitai) so any semi-realistic model should look similar I think.
@zakiry It was a test version of antlersmix_11. I must have forgot to tag it, fixed now. I didn't end up liking this mix so I didn't post it anywhere, but here most of the style comes from the loras (mainly dreamwave - it's on civitai) so any semi-realistic model should look similar I think.
I may as well ask because Iโm interested, but are these softwares free to use?
To begin with, these are not software. Simply put, LoRAs are models plus additional training data that you can create yourself and use to generate images. This image is generated using several LoRAs created by others. So, I have listed the LoRAs I used to generate this image in the previous comment. The website civitai, which I often use to find LoRAs, allows you to download these for free.
Only the pinky finger was inpainted as it was bent too much to the right but no manual edits at all! (The only manual edit in this whole pic was removing the background)
Your models are just something else, gotta love 'em.
Nice. Really catches the brisk and creepy autumn night that's appropriate for Halloween. Can just about hear those bats squealing and feel a chilly breeze.
Left eye look weird, but it looks a bit like ahegao so it's fine I think.
On another note, @ๆจๅ้ฒ้ธ if you've generated the images yourself, it'd be great if you could tag them as self_upload and you can add an artist tag for yourself. Links to loras could be placed in the commentary too (thanks for sharing the links btw, I wish more people did it for the less known loras).
Bottom of the horns are messed up as if there's extra piece of skin, and the teeth are inconsistent as well. I am generally not a fan of these animated ones since I've barely seen any that actually look good to my eye. It's all up to a preference of course, but hopefully you understand at least some of the downvotes now.
