AI Image Upscaler

Yes — drop up to 50 images at once. Pincel processes two at a time and queues the rest, so a batch of 30 typically wraps in a few minutes. Grab any result individually or pull the whole batch as a single zip.

Fast uses Google's upscaler — quick, cheap, and limited to 2× or 4×. Great for routine resolution bumps. Quality uses Crystal, which goes up to 10× and adds far more invented detail — better for small or noisy sources, prints, and anything where edges and texture matter. Quality is roughly 2× the cost of Fast at the same output size.

For most photos, 2× or 4× is the sweet spot. Scales beyond 6× really only pay off when the source is very small or you're targeting print. Each step costs more credits because the output grows quadratically — 4× makes the output four times bigger than 2×.

PNG is lossless — best when you'll edit further or need transparent backgrounds. JPG outputs at 95% quality, giving a much smaller file at near-identical visual quality. (Quality model only — Fast returns whichever format Google's upscaler produces.)

Fast is a flat 4 credits per image, regardless of size or 2×/4× scale. Quality scales with the output resolution in megapixels: 8 credits up to 4 MP, 16 up to 8, 32 up to 16, 64 up to 25, 128 up to 50, 256 up to 100, and 512 above that. Each card shows its exact cost upfront.

No — the AI doesn't just stretch pixels, it hallucinates new detail consistent with the source. Where a traditional resize would smear, Pincel adds plausible texture, edges and grain.

At 0 the model stays as faithful as possible to the source. Higher values (up to 10) let it invent more new detail — useful for very small or noisy inputs, at the cost of drifting further from the original.

Pincel handles inputs up to about 25 megapixels comfortably. With 4× scale, that produces 100+ MP outputs, which take longer to render. Each card shows the projected output size.

Yes — the upscaled images are yours to use commercially, including reselling and modifying.

Pincel does not store your inputs. They're only sent to the GPU server for processing, and discarded after.