Wildcard image generator with high-resolution refinement

The positive prompt supports seeded wildcard choices and is inserted into a style's {prompt} placeholder, or appended when no placeholder exists. The negative prompt is combined with the style's negative text. Rough width and height determine orientation, long-side size, and the nearest supported aspect ratio rather than exact pixel dimensions. The base generation uses CFG and negative conditioning for its first half, then positive-only guidance for its second half. The resulting latent is enlarged and refined using the shared CFG and separate refinement steps, denoise, sampler, and scheduler. Disabling latent resizing preserves refinement at the base size. Megapixel sizing and explicit aspect-ratio presets provide alternative output-size modes. The Pipe returns the refined image; the base-image preview is an intermediate result.

The positive prompt supports seeded wildcard choices and is inserted into a style's {prompt} placeholder, or appended when no placeholder exists. The negative prompt is combined with the style's negative text. Rough width and height determine orientation, long-side size, and the nearest supported aspect ratio rather than exact pixel dimensions. The base generation uses CFG and negative conditioning for its first half, then positive-only guidance for its second half. The resulting latent is enlarged and refined using the shared CFG and separate refinement steps, denoise, sampler, and scheduler. Disabling latent resizing preserves refinement at the base size. Megapixel sizing and explicit aspect-ratio presets provide alternative output-size modes. The Pipe returns the refined image; the base-image preview is an intermediate result.