What this checkpoint ischauhanOil4step is a full Stable Diffusion XL checkpoint, not a LoRA and not a style tag. It sits on theSDXL Lightning four-step line: a model distilled so that a short sampler trajectory already lands ona finished picture. On top of that speed, the weights have been aimed at one picture-kind. The kindis oil paint on canvas.I created this model by merging my Lora into it. (Based on SDXL 1 Base)

chauhanOil4step

What this checkpoint ischauhanOil4step is a full Stable Diffusion XL checkpoint, not a LoRA and not a style tag. It sits on theSDXL Lightning four-step line: a model distilled so that a short sampler trajectory already lands ona finished picture. On top of that speed, the weights have been aimed at one picture-kind. The kindis oil paint on canvas.Steps 4. A trial at 6 is allowed if a picture feels unfinished. Do not go to 20 to “improve” it.CFG / guidance 1.0 to 2.0. Start at 1.5. At 1.0 the prompt is quieter and the paint is freer. Above 3 the Lightning path breaks.Sampler Euler. Euler a is an acceptable second try. Avoid the long ancestral and SDE stacks on a first pass.Schedule Simple or Normal. SGM Uniform if listed — it matches many Lightning builds. Karras is a later experiment, not the default.Size 1024×1024. Portraits 832×1216 or 896×1152. Landscapes 1216×832 or 1152×896.Clip skip 1. If faces feel vague, try 2 once. Do not leave it at 2 out of habit from SD 1.5.VAE Automatic, if the VAE is baked. If midtones are grey, switch to the standard SDXL VAE.Seed Fix the seed while you change one word. Change the seed when the picture is right but you want another sitting.Batch Four images at the same short prompt is more useful than one image with a long prompt.

Files with SHA256: 7437b65e150d7863883fe5845975eb29fe5ca27eadaa669b358f804acf2c7af9