Future-Proofing the Working Artist: Catching Up on AI Without the Headache
If you are a working artist who has been trying to avoid AI so far, but you’ve finally realized you need to get a handle on this before you become completely extinct, then this article is for you. My hope with writing this is to help artists quickly jump into using AI productively without the steep learning curve of the myriad of apps available. I want to demystify the immense noise out there and elevate my fellow working artists to become invaluable in an ever-growing, competitive, and chaotic industry. But moreover, I want to show you how using AI in different ways can actually be immensely creatively rewarding.
The Noise
NanoBanana. Seedance. Veo3. Higgsfield. Stable Diffusion. Luma Ai. OpenArt Ai. Artlist. KlingAi. Midjourney. Flux. ChatGPT. Sora. Stable Diffusion.
What does all this mean?
For those of you who want a super quick answer to “what” in terms of how I make Ai work for me, here it is: I use Google Gemini’s NanoBanana Pro exclusively to create and rework images, and I use KlingAi to animate those images into video. I pay $28.34 / month for a Google One account, and $46.36 / month for KlingAi. This is currently all I need in my Ai toolset for working with images and video.
That’s it.
But for a more in depth look on the “why” and “how”, read on!
Making Sense of the Noise
Anytime you generate an image or a video, you are using an Ai model. Think of an Ai Model as a big digital brain that has studied millions of examples of specific things until it has become an expert at it. Each Ai model has been trained differently on different types of media – therefore each model has its own nuanced ‘flavor’. A year or so ago these differences were quite noticeable, but as technology advanced and models released newer versions, these differences have become less apparent.
AI models generally fall into two categories:
- Proprietary (Closed Source) Models
- Open Source Models
You will typically access these models in one of three ways:
- Installed locally on your computer (Open-source models only).
- Directly via the developer’s website or app (Proprietary models).
- Via a third party platform that bundles many different AI models together into their own unique interface (These are often called “AI Wrappers”).
Open Source models are free; you download them to your computer, access them through a specialized interface (called Comfyui, also free), and if your computer is powerful enough, gives you almost limitless customizations, free from any barriers or censorship. Stable Diffusion, SDXL, Flux, Wan, and Qwen are examples of open source models.
Proprietary Models are Ai models that are accessed via a company’s App or website, and you are charged either a monthly subscription fee or usage credits. Midjourney, Veo3, and Nano Banana are examples of these closed-source proprietary models.
Deciding between open source or closed source models is like the comparison between cooking at home and eating out at a restaurant. Open Source models are like cooking at home in a fully outfitted kitchen; your possibilities are limitless. Whereas proprietary models are like eating out at a restaurant, with quality varying between 5-star Michelin restaurants to McDonalds, but you typically only get what’s on the menu. You can still get a great meal eating out at a restaurant, but sometimes you may simply enjoy the process of cooking your own meal at home. Such with open source or closed source Ai models, both have their benefits and drawbacks.
Open Source Models
My journey into Ai started with learning how to use open source models. At the time, open source models were much more diverse and powerful than proprietary models (except maybe for Midjourney). Below is a fantastic tutorial video that explains how to get open source models up and running on your computer, as well as an introduction to what is possible. Its a looong video, but if you are just casually interested for now, skip through it a bit to get the gist of it.
Using open source models is a great solution if you dont want to spend any money, have at least 12 GB of VRAM, and enjoy the ability to have limitless technical customization.
For me, when I started working with open source models like Stable Diffusion (SD1.5, SDXL) and Flux, the biggest benefit was using the ControlNet module – this enabled me to use my own sketches or artwork to guide compositions, or use my art as style moodboards. I could generate an infinite number of images in my own art style (or any style), and bring my concept sketches to life in a fraction of the time, with near pixel-perfect accuracy.
Below is an example of how I started with a rough concept sketch, then used Stable Diffusion to produce a realistic version of it, then further developed it in Photoshop to realize my original vision. Without AI, I probably would’ve started by using stock photos of skeletons, a table, and chairs, then composited them together. Instead, AI essentially did this portion for me. Even though there was still quite a bit of work ahead of me in additional compositing and digital illustration in Photoshop, using AI saved me about 15 to 20 hours of work, fulfilled my original vision, and most likely contributed to a better final result.
Another benefit of using open source models is the ability to train your own models – this is how you can create a likeness of someone, an art style, or specific environment. This is called ‘LoRA’ training (short for Low-rank adaptation), where you create a smaller Ai model thats specifically trained on a specific set of images that is used with the foundational Ai model to generate your images. Using the cooking analogy – think of it as trying to make an Upside-down Pineapple Cake. A chef (your Ai model) generally knows how to make cakes, more or less. But to truly achieve an authentic pineapple cake, you will need to give the chef a specialized pineapple cake recipe (the LoRA) so they know exactly how to create one.
(There are also specially trained foundational models that are referred to as ‘fine-tuned checkpoints’, which, using the cooking analogy in this scenario could be considered as a specialized pastry chef, but I’ll skip over that in this article for the sake of simplicity).
Here is an example of LoRA training that I have done, with the consent of a model friend:
A lot of hobbyists make their own LoRAs based on styles, environments, cartoon characters, poses and actions, celebrities (without consent), and make them available for free on CivitAi. This is quite an ethically grey area, and CivitAi has recently cracked down on people uploading LoRAs based on copyrighted material or trained on real people without their consent (I believe the legal backlash against the viral Taylor Swift nude Ai images a while back played a big part in CivitAi’s policy changes). A word of warning to those visiting the Civitai website – it has some of the most depraved NSFW Ai images you’ll probably ever see, however it is one of the primary industry resources for fine-tuned checkpoint and LoRA models.
But here is the catch with open source models: building a LoRA and running ComfyUI requires a massive amount of technical tweaking, a high-end graphics card, and a lot of patience. It is the definition of a headache. Fortunately, a new generation of proprietary models has introduced a way to bypass this entirely…
Closed Source Proprietary Models
These are the AI models that you hear most about that are making the news. These aren’t free to download like open source models. These can only be used by creating an account with the provider, and paying a monthly fee or purchasing credits (however some offer limited free trial usage). The ones that are making waves now are:
Gemini – Nano Banana Pro (by Google):
Objectively the best image generator on the market for working professionals. I use this exclusively: generating polished stock images, text logos, bringing sketches to life, and applying moodboards to match my specific aesthetic perfectly.
Veo 3 (by Google):
One of the top video generators with great sound and speech capability. Through a Google One Account (through your Google Gemini interface) you can create up to 3 videos a day, or you can have increased access via different monthly plans through Google Flow.
(Here is where things can get a bit confusing – Google Flow via Google labs is a more professional-type interface where you can access Nano Banana as well as Veo video models. If you have a paid Google One Pro account for $28.34/month, you get access to 1000 monthly credits in Flow. To have access to more advanced features in Google Flow with 25,000 monthly credits, you’ll need a $169.99/month Google Ultra Plan. Personally I haven’t found this necessary for my needs).
KlingAi 3.0 (by Kuaishou):
Currently one of the best quality video generators. Capable of image generation as well. Has the ability to swap elements from videos – change characters, backgrounds, as well as lipsync function. You can also upload and create specific elements that will remain consistent across multi-shot sequences, or upload video to dictate camera movement and motion control. The ‘element creation function’ comes extremely close to replacing the LoRA training technique I explained earlier with open source models.
Seedance 2.0 (by Bytedance):
High quality video generation similar to KlingAi in function and capability. Some argue that it exceeds KlingAi’s capability in terms of cinematics – especially for action sequences. This is what news articles and social posts are talking about when they claim that “Hollywood is cooked.”
Midjourney (v7):
While earlier versions required you to use Discord (which confused a lot of non-tech artists), they now have a dedicated Web Interface. Midjourney is famous for its highly artistic, “fine art” outputs. It is a bit less focused on strict photorealism than Nano Banana, but it is popular for stylized illustration, concept art, and high-end aesthetics. A lot of the high-concept, scifi concept art that is floating around online for the last few years has most likely been made in Midjourney.
GPT-4+ (by OpenAi)
This is what we all just call ChatGPT. Its image generation function (which used to be known as DALL-E, but has since evolved into being fully integrated into ChatGPT) functions similar to Google’s Nano Banana. You can upload sketches or layouts to ask GPT to polish off into finalized illustrations, or you can endlessly edit and reiterate on images and graphics.
Grok Imagine (by xAi)
Elon Musk’s image and video generator has come a long way in the last 6 months. I don’t really feel like it’s suitable for any type of professional workflow, but it’s an easy tool to make fun, simple images and videos. Motion quality is very good, however prompt adherence still suffers compared to other models. Part of Grok Imagine’s popularity is that it is one of the only proprietary models that allows a certain degree of NSFW output.
Sora 2 (by OpenAi)
This AI model is currently geared towards creating meme-worthy videos, either 10 to 15 seconds long. Most of all the insane Ai vlogger videos or ridiculous ‘Cat waking up its owner’ videos you see on Tiktok are made in Sora. I dont see it part of any type of professional workflow anytime soon; I look at it as simply a fun diversion.
AI Wrappers
This is where I believe most of the confusion begins. You’ve probably seen hundreds of Ai apps online, all claiming to be the best at image generation, cinematic videos, exclusive access, or tout specialized functions for fashion brands, product photography, or anime. Dozens of websites pop up if you do a Google search for Nano Banana, Midjourney, or Seedance, adding to the confusion. This is because websites continually emerge offering bundled access to these models.
Think of it like your internet or cable provider offering you a bundled access package to Netflix, Prime, and Disney+, when you can also simply sign-up to them directly.
Common Ai Wrapper platforms you’ve probably seen are Higgsfield, OpenArt, Artlist, and Flora.
These platforms have their own unique interface to upload images, video, and text prompts, but they essentially just run the closed source models (and some open source models) I’ve listed earlier. Their main selling feature is that they give users access to many of the different leading models for a flat monthly fee or per credit packs. There is nothing more unique you can do on these platforms than you can already do by signing up directly to a Google One account for access to Nano Banana, or a KlingAi account.
Advantages to an AI Wrapper platform:
- Access to numerous models for a lower cost
- Industry-specific interface (eg: Flora for product and fashion photography)
- Good for experimenting with different models to see what you like best
Drawbacks:
- Stricter monthly caps on quality and quantity of image and video generations
- Inconsistencies and fluctuations in pricing
Putting AI to Work
As mentioned at the beginning of this article, all I typically use is Google’s Nano Banana for images, and KlingAi for video. My skeletons image I showed earlier was made about a year ago. At the time, creating it locally using an open source model via Comfyui was the optimal way to achieve that result. However, with the current capabilities of proprietary models, I could create it the same way within Nano Banana.
Here are some real world examples of how I currently use Ai in my workflow:
Logo / Illustration Design
Below is an example on how I used it to create a logo design for a hockey jersey. I used Nano Banana to develop some base concept designs, then brought into Photoshop for additional edits, before recreating as a vector file in Illustrator. In this way, Ai functions similarly to a junior designer on a project, speeding up the concept design phase.
Product Ad
This was a super quick way to showcase a product bundle. Without AI, I would’ve had to get all the pieces together for a photoshoot. Instead, I was able to develop this professional looking image in a matter of minutes.
Asset Creation
I was working on the design of a Christmas campaign for a hockey training facility. I wanted to use the typical holly leaves and berries as a design element, but wanted to make the leaves as green hockey sticks, and the red holly berries as red hockey pucks. I generated the asset images in a few tries, then composited them together while building the design in Photoshop. I was pretty jazzed with how it turned out haha.
Concept Art
I’ve been working on my own IP, Kayla From Earth, that started out as a comic book I created a few years ago. Nano Banana has been invaluable in bringing this to life. Below are some images of how I started with concept sketches, my comic book pages, and photoshopped character designs to create character sheets and cinematic keyframes, ready for animating into AI proof-of-concept shorts for pitches.
At the end of the day, AI isn’t here to replace your creativity; it is here to amplify it, if you let it. The landscape will keep changing, new models will drop, and the noise will continue. But as a working artist, your job isn’t to chase every single tech update. Your job is to find the tools that remove friction from your workflow so you can focus on what actually matters. Whether you are knocking out a quick product ad, designing a logo, or developing a massive sci-fi IP to eventually pitch for film and comics, these tools give you your time back. If you’ve been on the fence, there has never been a better time to jump in. Pick a ‘brain’ to hire, start experimenting, and see where it takes you.
If you ever have questions or just want some advice on getting started, always feel free to shoot me an email.