Snowy Hong Kong: Stable Diffusion Image Generation Practice

Translation note: This English version was translated by Codex (GPT-5) on 2026-04-20 18:01:46 CST. The source text is the corresponding Chinese post in this repository.

I first saw a snowy AI effect on the vivo X200 and wanted to recreate the same feeling on my own street photos, even though I did not have the phone itself.

1. Background

I wanted to generate a snow scene from ordinary Hong Kong street photography.

2. Technical Approach

My guess was that vivo used a model or local inference after taking the photo. I chose Stable Diffusion img2img as the practical workaround.

3. Execution

3.1 Installation

I downloaded a Stable Diffusion WebUI bundle made by the Bilibili creator “秋葉aaaki.”

Reference: https://www.bilibili.com/video/BV1iM4y1y7oA

3.2 Model Choice

I found a reference photo on Xiaohongshu and then located a matching model on Civitai:

https://civitai.com/models/109730/lwarchitecutralmix

3.3 Parameters

I used img2img with:

realisticVisionV60B1 v51HyperVAE.safetensors
vae-ft-mse-840000-ema-pruned.safetensors
ControlNet line-art mode
Default values for the remaining parameters

Prompt:

make it snow, snow scenery, modern architecture, glass, snowing, in winter, masterpiece, high quality, real, realistic, full detail, 8K, falling snow, there are no people or vehicles

Negative prompt:

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry

I kept the input image under 800×1200 to avoid out-of-memory errors.

Sampling settings: