Stable Diffusion v1.1: the original release, used creatively

Q: Everything is blurry

This is the model's nature. Lean into it or run the image through an upscaler and a newer model's img2img at low denoise.

Q: Prompt ignored

Weak text understanding. Shorten the prompt to the two or three most important nouns and raise CFG to 9-10.

Topic: Models

By Captain

Published 2026-09-29

Model card: sdv1

Open Model Info Open Hugging Bay

Stable Diffusion v1.1 is the first public Stable Diffusion checkpoint, trained at 256 then 512 pixels. This guide shows what it is good for today: dreamlike textures, abstract compositions and deliberately rough, early-AI aesthetics.

Overview

Stable Diffusion v1.1 from CompVis is the earliest released

checkpoint

of the series, trained first at 256x256 on LAION-2B and then at 512x512 on high-resolution LAION images. It is in the catalog as a piece of history and because its looseness is useful.

Expect soft forms, strange hands and heavy blur on people. Expect also beautiful abstract textures, painterly landscapes and a distinct 2022 look that newer models cannot fake.

Reference

Name	Type	What it is
Prompt	text	Short, evocative phrases. Long prompts are mostly ignored.
Negative prompt	text	Minimal effect; a few words at most.
CFG	7-10	Higher values force structure out of the noise.
Steps	30-50	It needs more steps than later models to settle.
Resolution	512x512	Stay square and small.

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

Write a single evocative line: a coral cathedral under green light.
Set 512x512,
CFG
8, 40 steps, Euler a.
Generate a batch of eight; variety between seeds is high.
Upscale a favorite with a modern
upscaler
to keep the texture but sharpen edges.
Optionally run the result through a newer model at 0.3-0.4 denoise to tidy forms while keeping the composition.

Examples

Abstract texture

a cathedral made of coral and bioluminescent algae, deep sea, green light, oil painting

Early-AI aesthetic

dreamlike photograph of a forgotten ballroom filling with fog, 2022 ai art style

Creative idea: album cover

album cover, a melting brass trumpet on a salt flat at dusk, grainy, surreal, cyanotype

Tips

Abstract and material prompts (molten glass, moss, rust, nebula) are where it shines.
Avoid faces unless you want the early-AI uncanny look on purpose.
Use it as a composition generator and refine in another model.
Film and print words (cyanotype, daguerreotype) suit its softness.

Troubleshooting

Everything is blurry

Why it happens

This is the model's nature.

How to fix it

Lean into it or run the image through an upscaler and a newer model's

img2img

at low denoise.

Prompt ignored

Why it happens

Weak text understanding.

How to fix it

Shorten the prompt to the two or three most important nouns and raise CFG to 9-10.

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.