Y
Prompt Workshop
Article · SD parameters

Stable Diffusion Hires Fix Guide

Hires fix (high-resolution fix) is the SD workflow that generates a small image then upscales and re-renders it. Choose the right upscaler and denoise strength and quality jumps noticeably.

Three key parameters

ParameterMeaningRecommended
Upscale ratioFinal size = base × ratio1.5×–2×
UpscalerInterpolation / AI scaler4x-UltraSharp / Latent
Denoising strengthDegree of regeneration after upscale0.3–0.5

Upscaler comparison

  • Latent — upscales in latent space, preserves style but occasional green blocks
  • 4x-UltraSharp — strongest for realism, recommended default
  • 4x-AnimeSharp — anime-specific
  • ESRGAN_4x — general, stable
  • R-ESRGAN 4x+ — best for photographic detail

Denoising strength

  • 0.2–0.3 — preserves composition, sharpening only
  • 0.4–0.5 — standard (recommended)
  • 0.6+ — changes details, can "grow new things"

Gotchas

4× upscale crashes VRAM

1024 × 4 = 4096 is too big for most GPUs. Stick to 1.5×–2×.

High denoise grows new faces

0.7+ regenerates face details — sometimes a different person.

Frequently asked questions

Does Flux have hires fix?

ComfyUI has equivalent latent-upscale + ksampler nodes. WebUI Forge supports it directly.

Is hires fix needed on SDXL?

Yes. Going from 1024 base to 1536–2048 still gives a clear quality lift.

Try this parameter live in the editor

The local editor includes Midjourney / SDXL / Flux templates with a visible, editable parameter panel.

Open the editor →
Yan · AI Prompt Workshop editorial team|Last updated on 2026-06-12。This site does not call any cloud model. Every prompt and parameter in this article was tested and refined locally by the editorial team.