This pipeline was finetuned from runwayml/stable-diffusion-v1-5 on the iamkaikai/amazing_logos_v3 dataset.
Latest Update: v3 model expands the training dataset to 200k imgages.

Training info

These are the key hyperparameters used during training:

Dataset size: 200k
Epochs: 7
Learning rate: 5e-07
Batch size: 1
Gradient accumulation steps: 2
Image resolution: 512
Mixed-precision: fp16

Prompt Format

The prompt format is as follows:

{template keywords} + [company name] + [concept & country] + [industry] + {template keywords}

For example:

Colored logo

Simple elegant logo for Google, G circle United states, technology, successful vibe, minimalist, thought-provoking, abstract, recognizable

monochrome logo (include "black and white")

Simple elegant logo for Google, G circle United states, technology, successful vibe, minimalist, thought-provoking, abstract, recognizable, black and white

Here are some examples of prompts:

Simple elegant logo for Digital Art, D A square, education, successful vibe, minimalist, thought-provoking, abstract, recognizable
Simple elegant logo for 3M Technology, 3 M square United states, technology and product, successful vibe, minimalist, thought-provoking, abstract, recognizable
Simple elegant logo for 42Studio, lines drop fire flame water, design studio, successful vibe, minimalist, thought provoking, abstract, recognizable, relatable, sharp, vector art, even edges

The [concept & country] section can include words such as:

lines
circles
triangles
dot
crosses
waves
square
letters (A-Z)
3D
Angled
Arrows
cube
Diamond
Hexagon
Loops
outline
ovals
rectangle
reflection
rings
round
semicircle
spiral
woven
stars

I invest in cloud GPU rentals to train my models. If you value what I do, consider buying me a coffee ☕️. Your support means a lot! https://www.buymeacoffee.com/iamkaikai666

iamkaikai/amazing-logos-v3

Text-to-image finetuning - iamkaikai/amazing-logos

Training info

Prompt Format

For example:

The [concept & country] section can include words such as:

iamkaikai/
amazing-logos-v3