High CFG amplifies its aesthetic bias. Drop CFG to 3-4 and add photographic words.

Trends

.ai

the best open source ai models

Playground v2 1024 Aesthetic: a second opinion to SDXL

Q: SDXL LoRA output is broken

Different weights despite the shared architecture. Lower the LoRA weight to 0.5 or skip LoRAs with this model.

Topic: Models

By Captain

Published 2026-10-02

Model card: playground2

Open Model Info

Playground v2 is an SDXL-architecture model trained from scratch for aesthetics. Learn how it differs from SDXL, its CFG and step preferences, and the painterly and realistic prompts where it beats the original.

Overview

Playground v2 (1024px Aesthetic) by Playground AI shares the

SDXL

architecture but was trained from scratch on a curated aesthetic dataset. Users preferred its images over SDXL's in Playground's own study, and it shows: colors are richer, compositions bolder and paintings more confident.

It is a good second opinion when SDXL gives you a bland result, handling realism, paintings and illustration alike.

Reference

Name	Type	What it is
Prompt	text	SDXL-style phrases; style and medium words are followed closely.
Negative prompt	text	Supported; keep it short.
CFG	3-7	Lower than SDXL; 4-5 gives natural tones.
Steps	25-50	30 with DPM++ 2M Karras or Euler.
Resolution	1024	Standard SDXL buckets.

No install required - run AI models on the cloud

BitVector Prism is the easiest way to start: pick a model, type a prompt and generate in a clean web app, with nothing to configure. BitVector is also available on Discord and on the web (SpyGlass).

Try BitVector

Step by step

Take the SDXL prompt you were unhappy with and run it here at
CFG
4.5, 30 steps.
Reduce quality boilerplate; Playground already aims for aesthetics.
Add a concrete palette (ochre, slate, cream) to make use of its color sense.
Try the same seed at CFG 3 and 6 to pick the mood.
Finish with a 1.5x hires pass at 0.3 denoise.

Examples

Painterly portrait

portrait of a flower seller at a night market, gouache painting, bold brush strokes, warm lantern light, ochre and indigo palette

Realism

photo of a vintage motorcycle in a desert garage, dust motes, late light through slats, 50mm, high detail

Creative idea: weather as character

a thunderstorm shaped like a galloping horse crossing a wheat field, dramatic painting, chiaroscuro, wide shot

Tips

It paints: oil, gouache and pastel prompts look especially good.
Portrait realism is strong at CFG 3-4 with lens words.
SDXL
LoRAs
partially work but often shift style; test before relying on them.
Dramatic sky and weather prompts are a strength.

Troubleshooting

Over-stylized faces

Why it happens

High CFG amplifies its aesthetic bias.

How to fix it

Drop CFG to 3-4 and add photographic words.

SDXL LoRA output is broken

Why it happens

Different weights despite the shared architecture.

How to fix it

Lower the LoRA weight to 0.5 or skip LoRAs with this model.

No install required - run AI models on the cloud

PirateDiffusion is Telegram only and built for pros: thousands of models, LoRAs and workflows driven by chat commands, with unlimited generation on a fixed price plan.