Wordclouds as Substack brand images?
Experimental new use for my personal NLP project from 4 years ago: generating Substack publication branding images
As a relative newbie to Substack, I’ve been digging into the ‘branding’ options offered for newsletters. Here’s a rundown of my past few days of experiments, in case it might be useful to others.
Substack’s image requirements
According to this post, 4 different image sizes with 3 different aspect ratios are needed for complete branding (all are optional, though):
256x256 (at least) - Logos for publications, publication sections, and personal profile, with transparent background, 1:1 aspect ratio
600x600 (at least) - Cover image, appears on publication welcome page (assumed to be used at 1:1 aspect ratio)
420x300 (minimum), 1456 x 1048 (recommended; can be larger) - Post image for search engine optimization, at 14:10 aspect ratio
1100x220 (or taller) - Email header image with a transparent background, aspect ratio 5:1 (will be less if taller)
A 5th optional image, with a different size and aspect ratio, is also possible: 1344×256 (at least) - a stylized “wordmark” for the header of the publication (minimum size, aspect ratio not to exceed 21:4).
So far I’m using the default wordmark on all but one of my publications, and that seems ok. Not having cover images on my welcome pages seems to be leaving out info for a potential reader, though, and I would like to have suitable images in the email headers for all of them.
My requirements
I want my publication branding images to be:
Meaningful for my publications and my readers
Low effort to create
Unique to me, and ethically generated
The same article's DIY Toolbox section has good advice on possible sources of images. I’ve been using some for images inside my posts. However, they don’t meet my third branding objective, being unique to me.
If you’re thinking at this point, just use an AI-based image generator to create something: I’m not comfortable with the ethics of doing that for my branding, given where things stand in the industry with image sourcing and rights.
💡A possible solution: wordclouds with unique, personal text inputs
While trying to figure out what kind of images to use for this publication and others I’ve started, I realized that perhaps I could use my personal wordcloud project from 4 years ago to generate a new wordcloud image for this newsletter. (I had used a wordcloud generated from my resume by the tool as my LinkedIn banner photo for a while.)
So I spun up the tool (with a bit of configuration work to get it running on my latest laptop with Python 3.11 and the latest packages), saved my LinkedIn profile to PDF to generate some text input, and ran the tool with a few built-in color combinations, word densities, and 2 square-ish shapes. Here are two of its outputs:
I wasn’t sure what to expect from my bio, because I had only tested the tool on resumes before. At first, I was surprised that analytics and AI didn’t stand out more. However, although AI & ML has been my primary focus for the past 10+ years, the tool doesn’t recognize chronology in the input; it’s just a bag of words. So the non-AI words later in the file (on earlier years of my career) are dominating. (Or maybe my LinkedIn profile needs a tune-up to balance what it says about me?) It made me so happy to see “team” emerge as a dominant word in every format, though 😊
Anyway, I’m leaning towards using the simpler circle format on the left for my newsletter cover photo. I may try different color combinations or densities, or a truncated version of the input file, though. (My tool UI already has 113 monochrome themes and 136 multicolor themes built in, and can support custom themes … lots of choices to try.)
Thoughts?
I’d love feedback on a few questions this experiment has raised:
Could a 600x600 (or 1200x1200?) wordcloud like this be a good choice for a newsletter cover photo?
If so, would either of these options I shared above ‘work’ for you as a reader to convey what my newsletter is about? What’s good and bad about them?
Could a 420x300 (or 860x600, or 1456 x 1048, …) wordcloud be useful for a post image / social sharing?
Could a 1100 x 220 (or taller) wordcloud be useful as an email header?
Could a 1344×256 wordcloud be useful as a wordmark image?
What do YOU think?
֎֎֎ Special free offer to thank you for reading / contributing ֎֎֎
Curious to see what your resume/bio/post wordcloud might look like?
For the first 20 people who ask (Comment or DM), I’ll generate a wordcloud set for you with my Python tool, and send it to you!
I’ll share screen shots of the tool’s current built-in color theme choices so you can pick your favorite(s)
You send me your text input - files can be .TXT, .RTF, .DOCX, or .PDF. (And yes, of course, any files you share with me for this purpose will stay 100% confidential and won’t be re-shared, ever).
Totally free, and you are welcome to use the wordcloud images however you like!
In the meantime, I’ve opened a new Github issue for myself to add these 3 custom Substack sizes to my tool’s UI:
1456 x 1048 (14:10)
1344 x 256 (21:4)
1100 x 220 (5:1), and maybe also 1100x440 or other taller options
The UI already supports the following sizes, plus a scaling factor to generate multiples of them, e.g. 600x600 or 900x900 circles:
Sometime in future, I’ll likely want to re-run the tool against my Substack post content, after I have more *new* posts here (the older posts migrated from Wordpress may not be representative), and see what pops up.
Credits:
My tool leverages amueller.github.io/word_cloud/ to generate the actual wordclouds
I built the tool’s UI with PySimpleGUI (v4 was free in 2020; I will need to migrate the tool to v5 during 2024Q2 under personal/hobbyist use only).
Just for fun, I ran the content of this post through my wordcloud tool. I was hoping to share it here on this post, but it doesn't seem that images can be attached in comments? So I've started a Chat and enabled images there - Subscribers can see it and comment :)