KirillM54

Промт не особо сложный, один на все рисунки:

dark shot, 80mm, Round, low hills, plowed up and sown to the top, scatter in broad waves; ravines overgrown with __plant_types__ wind between them; small groves are scattered in oblong islands; Narrow paths run from village. raw, hdr, professional photo , soft light, sharp, exposure blend, medium shot, bokeh, (hdr:1.3), high contrast, (cinematic:0.9), (muted colors, dim colors, soothing tones:1.15),(hyperdetailed:1.2), (noir:0.4)

Negative prompt: bw, grayscale, dof, neon, boring_e621_v4 easynegative, over saturated

Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 90607030, Size: 768x512, Model hash: aadddd3d75, Model: deliberate_v3, Denoising strength: 0.4, Hires upscale: 1.5, Hires steps: 12, Hires upscaler: 4x_NMKD-Siax_200k, Version: 1.5.1

Потом пропустил через Ultimate SD upscale чтобы размер на FullHD монитор можно было поставить. Добавил к себе в коллекцию сменяемых фонов рабочего стола.

Модель Can't believe it's not Photon с civitai. Изначальный промт на все рисунки:

hyper-realistic, highly detailed, highest quality, masterpiece, strange alien world, landscape full of life and color, distant stars, nebulas, sublime, breath-taking, high detail, sharp, cinematic lighting, 8k UHD __backgrounds_colors__ __tree_types__ <lora:FutureArchitecture:0.6> <lora:add_detail:1>, art by greg rutkowski and artgerm, soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed, (depth of field:1.4)

Negative prompt: (worst quality:1.4), (low quality:1.4), (normal quality:1.4), (low res:1.4), (deformed, distorted, disfigured:1.3), grainy, blurry, pixelated, noise, monochrome, animation, painting, drawing, illustration, anime, bad anatomy, bad hands, bad proportions, awkward posture, text, writing, numbers, watermark BadDream

Steps: 30, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Size: 768x512, Model hash: 87b4fe0b68, Model: cantBelieveItsNot_v10, Lora hashes: "FutureArchitecture: 0341597a84ba, add_detail: 7c6bad76eb54", TI hashes: "BadDream: 758aac443515", Version: 1.5.1

Темные варианты получились при использовании <lora:LowRA:0.5>. Еще в промежутке в img2img увеличил размер на 1,4 с Denoising strength: 0.45, а потом еще upscale х2 с помощью 4x_foolhardy_Remacri. Получилось как-то так.

Использую версию SD с DirectML под AMD (RX6600 8Gb). Хочется дождаться когда PyTorch прикрутят ROCm под Windows - должно ускорить генерации на AMD картах.