Prompt engineering is an art form. But sometimes, you have a vision in your head—or a folder of images on your hard drive—that words just can’t do justice.
What if you could stop trying to describe a “cyberpunk-baroque-oil-painting” and just show the AI what you mean?
I have developed two new ComfyUI workflows that change the way we interact with Flux, Wan, and other generative models. Instead of fighting with text, we are now using Visual Alchemy.
Here is the breakdown of the new tools dropping on Patreon today.
1. The “Vision Board” Generator (Free for All Patrons)
This is the foundational tool. We all have that one folder full of reference images—art styles we love, lighting setups we adore, or character vibes we want to replicate.
How it works:
You point this workflow at a folder of images. It creates a grid, feeds it to Google’s Gemini Vision AI, and asks a simple question: “What makes these images tick?”
It analyzes the composition, the color palette, the artistic medium, and the mood. Then, it writes a highly technical, director-level prompt for you. You don’t have to guess the keywords anymore. You provide the inspiration; the workflow provides the syntax.
The original input:
The original output:
2. The “Concept Injection” Protocol (Paid Members Only)
This is where things get… experimental. This is the Master Key to unlocking this concepts full potential.
While the first workflow replicates a style, this workflow fuses them. It allows you to take a “Base Reality” (e.g., a folder of cute, kawaii anime art) and inject a precise percentage of a “Foreign Concept” (e.g., visceral horror).
The “Kawaii Nightmare” Experiment:
I tested this by taking a folder of sugary sweet anime illustrations and injecting just 16% of a horror dataset.
The result wasn’t just a collage. The AI didn’t just paste a skeleton on top of a bunny. It synthesized the two concepts. It created a world where the bunnies were skeletons, where the blood was pastel pink, and where the horror felt strangely adorable.
It forces the AI to reconcile two incompatible ideas into a single, cohesive image. It is essentially a Visual Style Transfer Engine.
The output after the injection of 16% horror:
🏗️ The Engine Room: Google Gemini API
Both of these workflows rely on Google’s Gemini 2.0 Flash/Pro models to “see” your images.
“Do I have to pay for this?”
Technically, no. But realistically? Yes, and you should.
- The Free Route: You can use the free Gemini API key. However, it is rate-limited (approx. 20 calls/day depending on load), and because these workflows use 3 separate API calls per generation, you will hit that limit fast. Also, the free tier comes with heavy “Safety Filters.” If you try to generate anything remotely edgy (like our horror bunnies), the free API will likely block the request.
- The Paid Route: I strongly recommend attaching a payment method to your Google AI Studio account.
The Cost:
I have been stress-testing these workflows heavily. I have generated hundreds of prompts, run massive batches, and pushed the model to its limits.
Total cost so far? 1.22 SEK. (That is approx. $0.11 USD).
For the price of a tenth of a cup of coffee, you get faster speeds, higher rate limits, and—most importantly—no safety filters blocking your creativity. It is, in my opinion, the best value in AI right now.
📥 Get the Workflows
- Vision Board Generator: Available now for all Patreon tiers.
- Concept Injection (Master Key): Available now for Paid Patreon tiers.
NOTE: Both workflows require the Creepy Nodes pack, available at GitHub and in ComfyUI manager.
Stop typing. Start showing.
If you found this post useful, consider to sign up on my newsletter and get the latest news, tips and tricks directly in your inbox.
On my Patreon I regularly share exlusive materials, some are free for all members, and some are exclusive for paying members.