Ainavora Hires Fix: adult female-character portrait practice workflow
Article: https://ainavora.com/comfyui-hires-fix-guide/
Checked: 2026-10-05

Workflow: ainavora-hires-fix-woman.json
11 standard ComfyUI nodes; no custom node or dedicated upscale model.
The JSON has a checkpoint placeholder and contains no model weights.
Open it with Ctrl+O in ComfyUI, then select a compatible SD1.5 checkpoint.
It is not wired for SDXL or FLUX.

Example run: ComfyUI 0.33.4, isolated CPU mode, Windows.
Checkpoint: Realistic Vision V5.1 (VAE), full/fp32 SD1.5 model, by SG161222.
The model is not redistributed here; consult its own distribution terms.
Model author page: https://huggingface.co/SG161222/Realistic_Vision_V5.1_noVAE
File/version information: https://civitai.com/api/v1/model-versions/130072
The local checkpoint filename may be different from the distribution name.

Recipe: 512x512 first pass -> LatentUpscale bislerp to 768x768 -> second
KSampler -> decode and save. Both samplers use fixed seed 20261005,
20 steps, CFG 7, Euler, normal. The first pass denoise is 1.0; the JSON's
second pass default is 0.5. The article compares 0.25, 0.5 and 0.75.
The positive prompt describes a fully clothed adult woman by a window.
The sample is synthetic; no photograph or named real person was supplied.

The first Save Image stores a 512x512 baseline. The second stores 768x768.
Only change second-pass denoise and its Save Image prefix for a fair comparison.
Output, speed, and memory use vary with environment; no universal best value
or identity preservation is promised.

JSON and this README: MIT license, see LICENSE.txt. Model weights,
third-party software and generated images have separate rights.