Sign In

Aquarelle - A Vintage Anime Watercolor & Ink Style

1.1k

Download

1 variant available

fp16 SafeTensor

Aquarelle2.safetensors

Half precision, best balance • 435.35 MB

Verified:

Type

LoRA

Stats

112

1.1K

502

Reviews

Published

Jun 30, 2026

Base Model

Illustrious

Hash

AutoV2
708D1C856D

Trigger Words

aquarelle
default creator card background decoration
Followers - 391

391

Likes - 111

111

ComfyUI_02625_.png

A hand-painted illustration style LoRA drawn from the 1970s–80s anime and sci-fi fantasy era. Think ink line meets watercolor wash; loose, expressive, and unmistakably analog.

Aquarelle naturally renders characters with mature, expressive faces reflecting the full age range of its dataset. This is intentional. If you want younger results, add young to your positives. If you want older results, add young to your negatives and elderly, aged to your positives.

Trigger word: aquarelle - place it first, before everything else.

Recommended checkpoint: Ramthrust Pink Alchemy. Ramthrust preserves the ink linework and watercolor wash that define the style, while still pulling in a more modernized direction. You can play around with checkpoints, the style reacts differently to each. Comparison for my three recommendations below:

Other recommendations: DaSiWa's Epitaph Ecstacy, Janku Trained + Checkin & Noob AI + Rouwei Illustrious XL


Suggested Settings:

Euler A - Beta57 or Karras
CFG 5.0–5.2
28–30 steps
LoRA weight 0.85–1.0


Tips & Tricks:

1980s or 1970s in your prompt actively anchors the era and pulls the most authentic vintage illustration quality out of the LoRA. Highly recommended.

eye focus unlocks a surrealist mode - cosmic, introspective compositions with detailed eye imagery.

Short prompts outperform long ones. Five to eight tags is the sweet spot.

human, human face, human body in positives and animal, creature, beast, snout in negatives if you're getting unexpected non-human results. The LoRA reads descriptors very literally.

Aquarelle tends to get absorbed by other strong style LoRAs. It pairs best with character or concept LoRAs rather than competing style LoRAs.


Negative prompt:
worst quality, low quality, blurry, extra fingers, extra digits, malformed hands, bad anatomy, distorted face, deformed face, 3d, realistic, photorealistic, render, cgi, glossy, modern, contemporary


All gallery images generated straight from ComfyUI. No post processing, upscaling or inpainting (except where noted).