From sketch to scientific visual
Learn how to combine scientific accuracy and creative emotion using hand-drawn sketches restyled using Midjourney.
Generative AI easily makes “science-y-looking” things
With most general AI models, it takes almost no effort to prompt for a “25-year-old woman” or “a cat in a lab” and get photorealistic results in any lighting or mood.
What’s far more challenging is asking for specific cellular architecture like say, a neuron or a protein complex, and get a rendering that’s actually correct.
Most general models are good at hallucinating when it comes to scientific specifics. Molecules will have impossible bonds or neurons will look more like fantasy coral than real biology.
One fix is to build a custom library of scientific reference assets for your own model.
Another, and my personal favorite, is to start with a hand-drawn sketch based on verified references.
A hand-drawn sketch gives AI a scientific anchor or framework that guides enhancement. With a reference in place, a generative model can amplify off this anchor while maintaining relevant biological details.
In this article, I’ll show you how I do it.
Why good starting assets still matter in biotech marketing
AI has democratized visual storytelling, but it’s also made it easy to make wrong science look right.
In research, imaging tools like BioRender ensure precision.
For marketing, visuals are needed that not only instruct but also inspire.
Most biotech marketers don’t need to produce graphical abstracts for scientific publications, but they do need to create consistent brand assets that shape trust.
A campaign hero image or an emailer visual must make complex science feel accessible while staying true to the data behind it.
The right balance begins with an anchoring starting point.
My sketch-to-style method
I call my process “sketch-anchored AI styling.”
The workflow maintains creativity but ties back to scientific veracity while allowing tools like Midjourney, Photoshop, Canva, etc. to do what they do best, enhance and compose.
1. Sketch for structure
Start with an accurate reference.
For my sketches I use reference microscopy images, structural models, or anatomy diagrams to make sure proportions, details, and features are correct.
The lines don’t need to be perfect, just precise enough to capture relevant features. I start in pencil then outline in marker.
Next, scan or photograph the drawing under good light for a clear digital copy.
The following are two example sketches I made. A. a zygote B. a neuron
2. Use Midjourney moodboards and styles
I know there are many generative AI tools to create images. They all have their pluses and minuses.
Here I compiled a list of models popular with bioscience marketers followed by one key advantage and one key drawback:
🙂 Pro: Extremely rapid image generation (typically 3–5 seconds) and strong prompt fidelity.
❌ Con: Artistic depth and style variation are more limited compared to tools focused on creativity.
🙂 Pro: Exceptional in artistic expression and stylistic richness, making it ideal for creative concept images.
❌ Con: Less strong in text rendering accuracy and consistency for marketing assets requiring precision.
🙂 Pro: Easy conversational interface and natural language refinement of images, ideal for content creators and editors.
❌ Con: Offers fewer fine-tuning controls for style and artistic nuance, making it less suited when strong visual customization is needed.
🙂 Pro: Super accessible and integrated into a full-design platform — you can generate images, blend them with layouts, fonts, templates, and export visuals for web/email in one workflow.
❌ Con: It has limited tweaking controls and fewer credits for premium use, so when you need highly precise or scientific visuals, it can feel constraining.
🙂 Pro: Deeply integrated into the Adobe ecosystem (Photoshop, Illustrator, Creative Cloud) and trained on licensed/public-domain assets — which means your creations are commercially safe and you benefit from a smooth workflow.
❌ Con: Although the UI is friendly, it still struggles with very complex prompts or high-precision visuals (especially in specialized domains) and you may hit limits with resolution, credit allocations or fine-detail control.
My personal favorite model to style original sketches is Midjourney. I have written previously about how I use Midjourney to help with my creative process.
Take your doodles on a journey with Midjourney Omni-Reference
Imagination is the only limit to where your art can travel.medium.com
Use specific words to define style
For this project, I explored Midjourney’s style library using science-based keywords like the following:
- Dendritic — fractal, branching neuron or vascular forms.“dendritic patterning, fractal neural lattice”
- Bioluminescent — glowing organic light; ethereal illumination.“bioluminescent textures, glowing membrane lighting”
- Crystalline — geometric symmetry and faceted molecular detail.“crystalline microstructure, prismatic cellular formation”
- Helical — DNA-like twists and filament motion.“helical biomolecule, twisted organic architecture”
- Rugose — wrinkled, ridged, or textured surfaces.“rugose texture, corrugated biological layer”
- Translucent — soft light diffusion through organic material.“translucent cell film, diffused light microscopy”
- Fractal — recursive, repeating organic geometry.“fractal neuron branching, recursive cellular form”
- Collagenous — fibrous, connective-tissue structures.“collagenous fiber mesh, fibrous bioarchitecture”
- Morphogenesis — visual growth, emergence, or organization.“morphogenesis in motion, emergent cellular structure”
- Quaternary — multi-unit protein assembly and symmetry.
“quaternary protein complex, modular folding pattern”
During my research, I also found several useful sref (style reference) codes, which I combined with Midjourney’s Retexture feature to restyle my hand-drawn sketches into polished, science-inspired visuals.
The following images show my zygote restyled with Midjourney sref codes.
You can see Midjourney kept the structural details of my zygote intact but restyled with amazing finishes.
When you generate a number of images that have the finishes you want based on color and texture, pop them into a Midjourney moodboard to go even more granular with a style.
Here is my neuron sketch restyled with a cellular moodboard I created.
These images are great on their own, but a key to making good science collateral is additional layering and composition.
3. Merge art and science
Once you’ve finished restyling your original sketches in Midjourney, open Photoshop or Canva to layer those elements into a cohesive composition.
Here’s an example: a composite image of my restyled neuron, placed on an abstract background generated with an sref code from my science-based Midjourney style search.
Here is another composite image with restyled zygote.
Each effect should clarify structure while also inspiring connection.
The best part is that the results all stay grounded by the original sketch, the anchor of scientific veracity.
4. Adjust color to communicate emotion or brand
Color isn’t just a decoration. Color adds an important narrative.
From an emotional standpoint, colors have meaning. For example:
- Blue-greens = trust, precision
- Golds and corals = innovation, discovery
- Violets = molecular energy and creativity
Choose tones that express both brand mood and biological realism.
For hero banners, I might highlight fluorescence microscopy gradients or for print it is important to match brand Pantones.
Don’t forget to relay sciences stories using the right colors.
5. Verify before publishing
One thing I find disturbing with biotech companies turning all their marketing over to AI agents is the lack of oversight on the content that is generated.
I always crosscheck final outputs with original references to confirm that lighting or stylization hasn’t distorted structural truth.
Accuracy builds trust but emotion is retained in memory.
Even subtle inaccuracies can erode confidence in a brand. A final check to review accuracy of AI-styled output is what separates scientific storytelling from science fiction.
Lessons from AI hallucinations
In 2024, two scientific papers were retracted because AI-generated figures depicted impossible anatomy. One paper displayed a rat with huge reproductive organs and another paper a human leg with the wrong number of bones.
If AI can misrender a femur, it can misrender much more obscure scientific elements.
Starting your biotech design with a confirmed reference like a qualified sketch is one way to create a scientific guardrail to keep AI from hallucinating something fictional.
AI image generators are great tools, but they are not great at oversight.
Why accuracy matters for biotech brands
Science-driven companies compete not only on data, but on trust.
When visuals are both accurate and moving, they bridge the gap between research and recognition.
A hand-drawn base image gives an AI image generator a map. The resulting imagery captures not only what a technology does but what it feels like to innovate.
- Start with accuracy.
- Layer in atmosphere.
- End with emotion.
Combining accuracy with emotion is how a biotechnology brand can move forward from being just credible to becoming infinitely memorable.
Anchoring the process
There’s something gratifying in creating a hand-drawn sketch, specifically to connect with biology.
Each line connects back to real biological structures that are important for function.
I like to think of using AI tools in my process as amplifiers not true inventors. Making a shift from viewing AI as a sole generator to true collaborator is what makes human-led AI art so much more powerful.
Because emotion without truth is fiction and truth without emotion is forgettable.
TL;DR Biotech Assets Summary
🧪 Sketch first to define accuracy
🧪 Use AI more to explore style, not structure
🧪 Colorize with emotional intent or for brand
🧪 Always verify content before publishing
I like to design biotech visuals that connect with both sides of the mind: accurate enough for scientists but evocative enough to move everyone else.
Where this fits in scientific storytelling
This approach reflects how I work across scientific content—combining structure, visual thinking, and modern tools to shape ideas into something people can understand and use.
The goal is not just to create images, but to build visual systems that support communication.
Need help shaping your scientific visuals?
If you’re working with complex material and need clarity across visuals, presentations, or technical documents, this is the kind of structure I bring into projects.