lobelia-sdxl

For best outputs and usage:

• Sampler: Euler a, Euler, DPM++ 2M Karras, DPM++ SDE

• Steps: 40, cfg 7.5, CLIP skip 1.

• VAE: sdxl_vae.safetensors.

• Usage: this is a versatile type of model, which means it can do both cartoon / realism. In order to switch, play with the pos and neg prompts, and switch samplers as follows:

  • Realism:

    • Pos prompt: for example, start your pos with "Cinematic, epic, professional photography, maximum hyperrealism, photorealism", then your text, then end it with "16K, high quality, photorealistic, realism, art photography, Nikon D850, 16k, sharp focus, masterpiece, breathtaking, atmospheric perspective, diffusion, pore correlation, skin imperfections, DSLR, 80mm Sigma f2, depth of field, intricate natural lighting, film grain, background Cinematic Hollywood Film Style, shallow depth of field, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy".

    • Neg prompt: "cartoon, anime, render, cg, 3D, desaturated, text, watermark, low-quality, signature, moir pattern, downsampling, aliasing, distorted, blurry, glossy, blur, jpeg artifacts, compression artifacts, poorly drawn, low-resolution, bad, distortion, twisted, excessive, exaggerated pose, exaggerated limbs, grainy, symmetrical, duplicate, error, pattern, beginner, pixelated, fake, hyper, glitch, overexposed, high-contrast, bad-contrast, epiCNegative", being the embedding "epiCNegative" optional (included in the realistic example runs).

    • Sampler: DPM++ 2M Karras, DPM++ SDE

  • Cartoon, anime, other:

    • Pos prompt: "(cartoon, 2d, flat 2d style)," at the beginning; not pointing this / the realism keywords tends to create anime-like outputs by default.

    • Neg prompt: "desaturated, text, watermark, low-quality, signature, moir pattern, downsampling, aliasing, distorted, blurry, glossy, blur, jpeg artifacts, compression artifacts, poorly drawn, low-resolution, bad, distortion, twisted, excessive, exaggerated pose, exaggerated limbs, grainy, symmetrical, duplicate, error, pattern, beginner, pixelated, fake, hyper, glitch, overexposed, high-contrast, bad-contrast, EasyNegativeV2, unaestheticXL2v10", being the "EasyNegativeV2" and "unaestheticXL2v10" embeddings included in the cartoon example runs.

    • Sampler: Euler a, Euler