AI for Artists

Start Here

This course is written for people who make things — illustrators, designers, photographers, art directors, animators, musicians, hobbyists. It assumes you want to *use* these tools well, not build them. There is no maths here, no code, and nothing you need to install.

Everything is organised around four questions: what actually happens when you type a description, how to write descriptions that get you what you pictured, which tool to reach for on a given job, and what you're allowed to do with the result once you have it.

What you'll be able to do by the end

Explain, in plain language, what the machine is doing between your prompt and your picture — and use that to debug bad results instead of guessing
Write prompts in whichever of the two prompting dialects your chosen tool speaks
Pick the right tool for a job in under a minute, and know roughly what it will cost you
Steer a result with reference images, masks and edits rather than rerolling the dice
Answer a client asking "do I own this, and can we ship it?"

What is deliberately not here

Model architectures, U-Nets, latent spaces, LoRA training, ComfyUI graphs, VRAM budgets and local installation. None of that makes you a better artist, and all of it is a distraction while you're learning to see.

If you later want it — and some of you will — it lives in Deep Learning &rsaquo; Diffusion Models, where it belongs alongside the rest of the theory.

One thing to hold onto before you start

These tools do not understand your idea. They have seen an enormous number of images with words attached, and they are very good at producing something statistically consistent with the words you gave them. Every technique in this course is really the same technique: give it words and references that narrow down what it could plausibly produce until the only plausible answers are the ones you'd be happy with.

What Actually Happens When You Hit Generate

Almost every frustration artists have with these tools — the ignored instruction, the sixth finger, the four petals when you asked for five, the two wildly different images from the same prompt — comes from one thing: not knowing what the machine is doing with your sentence.

So before any prompting advice, here is the whole journey your description takes. It is four steps, and none of them require any technical background.


  you type a description

        |

  1. a language model reads it       -> turns words into meaning

        |

  2. the meaning becomes a target    -> a 'this is what good looks like' signal

        |

  3. a painter starts from static    -> and refines toward that target, 20-50 times

        |

  4. the result is decoded to pixels -> your image

Step 1 — A Language Model Reads Your Words

The first thing that touches your prompt is a language model. This is the part of the system people mean when they say "LLM," and it is worth understanding because it is the part you actually talk to.

What a language model is, for an artist

A language model is a program that has read a colossal amount of text and learned which words tend to go with which other words, in which contexts. That's genuinely it. From that one ability it gets a surprisingly deep working knowledge of what things mean — that cerulean is a blue, that chiaroscuro implies a hard light source and deep shadow, that a Dutch angle is a tilted camera, that Bauhaus and Art Nouveau are not the same century.

It is not looking anything up. It has no dictionary. It has a learned sense of how these words sit next to each other, and that turns out to be enough.

What it does with your prompt

It converts your sentence into a compact numerical summary of its meaning — not a list of your words, but a representation of what you were getting at. Two prompts that mean nearly the same thing land in nearly the same place; two that mean different things land far apart.

This is why "a decrepit seaside hotel" and "a run-down hotel by the sea" give you similar images, while changing one adjective from decrepit to opulent changes everything.

Two flavours you'll meet in the wild

A plain text encoder. Older and open-weight image tools use a relatively small model that just reads your words. It is literal. It does not reason, it does not fill gaps, and it does not ask what you meant. Give it a vague prompt and you get a vague image.
A full conversational model. Tools built on chat assistants (ChatGPT's image generation, Gemini's, Copilot's) run your request through a proper reasoning model first. It will expand "a cosy reading nook" into a paragraph of specifics before any painting starts. Convenient, but it means the model is making creative decisions on your behalf — which is why the same prompt twice can give you two quite different rooms.

Knowing which of these you're talking to is the single most useful thing you can know about a tool. It decides how you should write.

Step 2 — Your Words Become a Target

That numerical summary of your meaning now becomes a target — a standing instruction that will be consulted at every single step of the painting process.

The useful mental picture is an art director standing behind the painter, holding your brief, and at every stage saying "more like this, less like that."

The director does not paint. It only judges direction. And critically, it judges the whole picture at once against the whole brief at once — it does not tick off your requirements one by one.

That single fact explains a lot of behaviour you have probably found maddening, and we'll come back to it two sections from now.

How hard the director pushes is usually a setting you control. Different tools call it different things — guidance, prompt strength, stylize, CFG, creativity — but it is always the same dial:

Push too gently → the model wanders off and makes something pretty that isn't what you asked for
Push too hard → the model over-corrects: scorched colours, crunchy contrast, stiff and literal compositions
The sweet spot is usually near the default. When results look burnt or over-saturated, this dial is the first thing to turn down, not your prompt.

Step 3 — The Picture Is Refined Out of Static

Here is the part that surprises people. The model does not draw your image. It does not start with a blank canvas and add a horizon, then a boat, then a sky. It starts with a rectangle of pure random static — visual noise, like an untuned television — and repeatedly *removes* the parts that don't belong.

The sculpture analogy is the accurate one. The model is not painting; it is carving. It looks at the noise and asks "if this static were a slightly blurry version of the image described in the brief, what would I need to take away?" It removes a bit. It looks again. It removes a bit more.

Do that twenty to fifty times, checking against your brief every time, and a coherent image emerges from the static. Early steps decide the big things — composition, masses, where the light comes from. Late steps decide texture, edges, fine detail.

Why this is worth knowing:

The composition is locked in early. By roughly a quarter of the way through, the layout is decided. This is why prompt words about composition and framing carry more weight than words about small details — and why adding "tiny brass buttons" rarely gets you tiny brass buttons.
More steps is not more quality. Past a point the image stops changing meaningfully and you're just spending money. Defaults are almost always fine; doubling the steps is not a fix for a weak prompt.
The starting static is a real ingredient. That random rectangle is generated from a number called the seed. Same prompt + same seed + same settings = the same image, every time. Same prompt + different seed = a genuinely different image. When you get something almost right, locking the seed and changing one word is how you make a controlled adjustment instead of rolling the dice again.

Step 4 — And Then: Pixels

For speed, most of that carving happens on a small, compressed sketch of the image rather than on the full-resolution picture — think of it as working on a thumbnail. A final decoding step expands that sketch into the actual pixels you download.

You never see this step, but it leaves fingerprints, and recognising them saves you from chasing prompt fixes for problems your prompt didn't cause:

Small faces and hands in wide shots come out mushy. In the compressed sketch a distant face is only a few marks wide; there simply wasn't room for the detail. The fix is composition or a face-aware upscale — never a longer prompt.
Fine repeating patterns wobble. Chain-link, houndstooth, brickwork, crowd scenes, distant text.
This is the natural resolution ceiling. Every tool has a size it was built for. Ask for something far larger in one shot and you get duplicated limbs and repeated horizons. Generate at the native size, then upscale.

What This Explains About Your Results

Now the payoff. Nearly every classic complaint about AI image tools is a direct, predictable consequence of the four steps above — which means each has a real fix rather than a superstition.

What you see	Why it happens	What actually fixes it
"No hats" produced a hat	The brief is matched as a whole, not read as an instruction list. Hat is in the brief, so hattishness scores well. The director never learned the word "no".	Describe the positive: bare-headed, windswept hair. Or use the tool's dedicated negative-prompt field, which is a separate mechanism from your sentence.
You asked for five, got four	Nothing in the process counts. It matches the look of a described scene, and five petals and four petals look almost identical to a whole-image judgement.	Ask for small numbers, or arrangements it has seen often (a pair, a row of three). For exact counts, generate then edit.
Hands and teeth	Hands appear in a million poses, from a million angles, half-occluded. There is no single "look of a hand" to converge on — unlike a face, which is always roughly the same arrangement.	Genuinely much better in 2026 models. Otherwise: crop them out, or inpaint them afterwards.
It merged two subjects	The brief is one summary of one meaning. "A woman in a red coat and a man in a blue coat" is, to that summary, largely people-in-coloured-coats. Attributes bleed.	Separate them in space (on the left… on the right…), or generate separately and compose. Some tools support per-region prompting.
The style words did nothing	A style name only works if that style was strongly represented, and named that way, in what the model learned. Obscure or very recent artists often simply aren't in there.	Describe the style's mechanics instead of naming it: the medium, the mark-making, the palette, the light, the era. This works on every model.
Same prompt, wildly different images	Different starting static — and, on chat-based tools, a reasoning model reinterpreting your request afresh each time.	Lock the seed. On chat tools, be specific enough that there is nothing left to reinterpret.
Text came out as gibberish	The model carves shapes, and letterforms are shapes that have to be exactly right to read as language.	Use a model built for it (Ideogram, Qwen-Image, GPT Image, Recraft), keep the text short, and put it in quotation marks.

The general principle: when a result is wrong, ask which of the four steps failed? Wrong subject or wrong meaning is step 1 — rewrite. Right subject but ignored or over-cooked is step 2 — adjust guidance. Wrong composition is step 3 — reroll or change the seed. Mushy detail is step 4 — recompose or upscale. That single question will save you more time than any prompt template.

The Same Idea in Video, 3D and Sound

Once you have the four steps, the other media are small variations rather than new subjects. You do not need to learn them from scratch.

Video does the same carving, but on a stack of frames at once rather than one picture — which is why a model can keep a subject consistent as it moves. It is also why video is expensive, why clips are short, and why the thing to describe is motion and camera ("slow push in", "she turns to look over her shoulder") rather than a static scene. A described scene with no described motion gets you a slow drift.
3D mostly works by generating a subject from many viewpoints and reconciling them into a solid shape with a surface. That reconciling stage is why AI 3D output needs cleanup: the geometry is plausible from outside but rarely tidy underneath.
Music and sound carve a picture of the sound over time — effectively a spectrogram — then convert it to audio. Same process, different canvas. Which is why music tools respond to the same kind of description you'd give an image: genre, instrumentation, mood, tempo, production era.

The practical upshot: everything you learn about prompting images transfers. Be concrete, describe what can be perceived, put the important thing first, and change one variable at a time.

Writing Prompts That Work

Prompting is not a secret language and there is no magic word. It is art direction in writing: saying enough that a competent stranger could produce roughly what you pictured, and no more.

The good news is that the fundamentals are the same across every tool and have been for years. Only the dialect changes.

The Six-Part Recipe

Nearly every prompt that works contains the same six ingredients. Not every prompt needs all six — but when a result feels generic, it is almost always because one of these was left for the model to decide.

Subject — what it is, specifically. A heron, not a bird.
Action or state — what it is doing. Mid-stride through shallow water, not just standing there.
Setting — where, and when. A tidal estuary at first light.
Medium or style — what kind of image this is. 35mm photograph, gouache illustration, charcoal study.
Light — the single highest-leverage word group in the whole prompt. Low backlight, mist catching the sun.
Framing — where the camera is. Low angle, long lens, subject small in frame.

Watch it build:

a heron → a stock bird on a white background
a heron mid-stride through shallow water → now it's a scene
…in a tidal estuary at first light → now it has a world
…35mm photograph → now it has a medium, and stops looking like a render
…low backlight, mist catching the sun → now it has mood, and this is where most of the quality arrives
…low angle, long lens, subject small in frame → now it has a point of view, and it looks authored

Put the subject first. Most models weight the front of a prompt more heavily. Whatever the image is about should be in the first handful of words.

Cut the adjectives that describe quality rather than content. Beautiful, stunning, masterpiece, 8k, highly detailed, award-winning were load-bearing in 2022 and are mostly noise in 2026 models — they crowd out the words that are actually doing work. If a word doesn't change what a viewer would see, delete it.

The Two Dialects: Sentences or Tags

There are exactly two ways to write a prompt, and using the wrong one for your tool is the most common reason a good prompt underperforms. Find out which dialect your tool speaks before anything else.

Dialect 1 — Natural sentences. Write as you would brief a person. Used by GPT Image, Gemini/Nano Banana, Midjourney, Ideogram, FLUX, Runway, and essentially every tool built on a conversational model.

A low-angle 35mm photograph of a heron mid-stride through a shallow tidal estuary at first light, low backlight with mist catching the sun, the bird small in a wide frame.

Full sentences genuinely help here — the language model uses grammar to work out what modifies what. "A red car and a blue house" survives as a sentence; as tags it becomes a colour soup.

Dialect 2 — Comma-separated tags. Short descriptors, most important first. Used by Stable Diffusion and its descendants, and by most open-weight and anime-focused models.

heron, mid-stride, shallow estuary water, first light, backlit, mist, 35mm photo, low angle, wide shot

These models read more literally and weight roughly by position, so ordering is your emphasis. Many also accept explicit weights and a separate negative prompt field for things to avoid — which, unlike writing "no X" in your sentence, actually works.

A quick way to tell which you're in: if the tool has a chat box and answers you back, it's dialect 1. If it has a second text box labelled negative prompt, it's dialect 2. If in doubt, write a natural sentence — every modern model tolerates it, while tag soup confuses the conversational ones.

Platform-specific extras are worth ten minutes of reading. Aspect ratio, stylisation, reference weighting and version flags are all per-platform, they change every few months, and every vendor documents them properly. Read your tool's own guide once; skip anyone's list of "magic prompt words".

Describing What You See

The skill underneath prompting is not writing. It is looking — being able to name what makes an image you admire work. Artists who develop this vocabulary get better results from every tool, permanently, because it is the vocabulary the models were trained on.

Light — the highest-leverage category, and the one beginners underuse most.

Direction: backlit, side-lit, top light, underlit
Quality: hard shadows, soft diffused light, overcast, dappled
Time and source: golden hour, blue hour, noon sun, candlelight, sodium streetlight, screen glow
Contrast: high-key (bright, low contrast), low-key (dark, dominated by shadow), chiaroscuro

Camera and framing — this is how you control composition.

Shot size: extreme close-up, portrait, medium shot, wide establishing shot
Angle: eye level, low angle, high angle, overhead, Dutch angle
Lens behaviour: wide angle (expansive, distorted edges), long lens (compressed, shallow focus), macro, fisheye
Depth: shallow depth of field, everything in focus, foreground framing

Colour — describe a relationship, not just a hue.

Muted earth tones, high-saturation neon, monochrome with a single red accent
Complementary orange and teal, analogous blues and greens, desaturated with warm highlights

Surface and material — what makes an image feel touchable.

Weathered, polished, matte, translucent, wet, dusty, hand-worn
For illustration: visible brushwork, heavy grain, clean vector edges, rough ink line

A practical drill. Take three images you wish you had made. For each, write six lines: subject, light, colour, framing, medium, and mood. Do that for a week and your prompting improves more than any template will manage. You are building the vocabulary; the tools already speak it.

Getting the Style You Want

Style is where most artists start and where most get stuck, because the obvious move — naming an artist — is both the least reliable technique and the most ethically loaded one.

Describe the mechanics instead of naming the style. This is the technique that works on every model, never goes stale, and gives you something nobody else has. Instead of in the style of [artist], specify what actually produces that look:

Medium and tool: gouache on toned paper, ballpoint pen, linocut, 35mm colour negative, wet-plate collodion
Mark-making: visible brush texture, hard-edged flat colour, cross-hatching, heavy impasto, airbrushed gradients
Palette and light, as in the previous section
Era and context: 1970s sci-fi paperback cover, mid-century travel poster, 1990s film still, Edo-period woodblock

Four such phrases will out-perform an artist's name almost every time, and the result will be yours.

Where naming still helps: movements and genres (Art Nouveau, Brutalist, Ukiyo-e, Bauhaus, cyberpunk), which are broad, well-documented and not attached to a living person's livelihood.

Where it fails: individual living artists — often absent from the model, unpredictable when present, restricted or blocked on several major platforms, and a genuine ethical problem when the artist is alive and working. Most agencies and clients now have a policy on this; if you work commercially, ask before you use it.

Style references beat style words. Nearly every 2026 tool accepts an image as a style reference, and passing your own work, your own photographs, or your own mood board is more precise than any sentence, more defensible than an artist's name, and yields a look that's consistently yours. If you take one thing from this section: build a reference folder before you build a prompt library.

Iterating: From First Try to Final

The single biggest difference between people who get great results and people who don't is not prompt quality. It is that one group changes one thing at a time, and the other group rewrites the whole prompt after every disappointing result and learns nothing.

A working loop:

Start short. Subject, setting, medium. Four or five generations. You are looking for a direction, not a final image.
Pick the closest one and lock its seed. Now you have a stable base and every change you make is legible.
Change one thing. The light. Then the framing. Then the palette. Note what each word actually did — this is how you build real intuition rather than folklore.
Switch from generating to editing. Once composition is right, stop rerolling. Inpaint the problem areas, extend the canvas, adjust by conversation. Rerolling a good image to fix a small flaw usually loses more than it gains.
Finish outside the tool. Upscale, then colour-grade, crop and retouch in whatever you normally use. Almost no professional output ships straight from the generator.

Habits worth having:

Keep a prompt journal. Prompt, seed, tool, settings, and one line on what worked. This is also your evidence of authorship if you ever need to prove it — see the section on copyright.
A/B a single variable. Two generations differing by one phrase teach you more than twenty random ones.
Reroll before you rewrite. If a prompt gives one good result in five, the prompt is fine and you were unlucky. Judge a prompt on four generations, not one.
Know when to stop prompting. Some things are faster to fix in Photoshop in thirty seconds than to coax out of a model in thirty minutes. Prompting is a tool in your workflow, not the whole of it.

Choosing Your Tool

There is no best tool, and anyone who tells you otherwise is selling something. There are tools that are clearly better at specific jobs, and the useful skill is matching the job to the tool in under a minute.

Everything in this section runs in a browser. You do not need a powerful computer, a graphics card, or an installation. If you have a laptop and a card on file, you can use all of it today.

Pick by the job, not by the brand:

If the job is…	Reach for
A finished, realistic image you'll actually publish	GPT Image, FLUX.2
Art direction, mood, cinematic concept work	Midjourney
Anything with readable text in it — posters, packaging, logos	Ideogram, Qwen-Image, Recraft
Brand and design assets, vector, consistent style sets	Recraft
Editing an existing image by describing the change	Gemini / Nano Banana Pro, GPT Image
Cleared-for-commercial-use assets inside Creative Cloud	Adobe Firefly
Cinematic video with sound	Veo, Sora
Lots of video on a budget	Kling, Hailuo
Precise camera control and video editing	Runway
Game-ready or print-ready 3D	Meshy, Tripo
Music you can legally release	ElevenLabs Music, Soundraw
Trying many models without many subscriptions	OpenArt, Freepik, Krea

How the money works, briefly. Almost everything is credits or a monthly tier. A still image is cents; a few seconds of video is dollars — video is roughly two orders of magnitude more expensive than stills, which is why you should storyboard with images first and only generate video once you know the shot.

Before you subscribe to anything, spend a week on an aggregator (below) where a single subscription gets you most of the major models. You will discover which two you actually reach for, and you can then subscribe directly to those.

A word on how fast this moves. Everything named on this page will have shifted within a year — models get versioned, renamed, deprecated and occasionally shut down mid-project. Treat these as categories with current examples, not a shopping list. What doesn't change is the six-part recipe, the two dialects, and knowing what the machine is doing.

Images

The 2026 image landscape has settled into clear specialisms. The blunt version: GPT Image for realism and reliable prompt-following, Midjourney for art direction, FLUX when you want open weights or licensing flexibility, Ideogram or Qwen-Image the moment there is text in the picture, Recraft for brand and design systems, Firefly when the client needs indemnified assets.

GPT Image (OpenAI)

The strongest all-round default in 2026 and the top of most blind-preference leaderboards. Excellent realism, unusually literal prompt-following, solid text rendering, and conversational multi-turn editing — generate, then say 'make the sofa navy, pull back a little' and it does. If you only learn one tool, this is the safe choice.

AI for Artists

Start Here

What Actually Happens When You Hit Generate

Step 1 — A Language Model Reads Your Words

Step 2 — Your Words Become a Target

Step 3 — The Picture Is Refined Out of Static

Step 4 — And Then: Pixels

What This Explains About Your Results

The Same Idea in Video, 3D and Sound

Writing Prompts That Work

The Six-Part Recipe

The Two Dialects: Sentences or Tags

Describing What You See

Getting the Style You Want

Iterating: From First Try to Final

Choosing Your Tool

Images

GPT Image (OpenAI)

Midjourney

FLUX (Black Forest Labs)

Google Nano Banana Pro / Gemini Image

Ideogram

Recraft

Qwen-Image (Alibaba)

Adobe Firefly

Leonardo AI

Video

Google Veo

Sora (OpenAI)

Kling

Runway

Luma Dream Machine

Pika

Hailuo (MiniMax)

3D and Assets

Meshy

Tripo AI

Rodin (Hyper3D)

Luma AI

Tencent Hunyuan3D

Music and Sound

ElevenLabs Music

Suno

Soundraw

Mubert

AIVA

All-in-One Playgrounds

OpenArt

Krea

Freepik

Hugging Face

Controlling the Result

Reference Images: Show, Don't Just Tell

Keeping a Character Consistent

Editing by Conversation

Inpainting and Outpainting

Image-to-Image

Guided Generation: Pose, Depth and Line

Upscaling and Finishing

Working Professionally

Can You Copyright It?

Can You Sell It? Licensing in Practice

Training Data and Choosing Ethically

Building a Repeatable Workflow

Where to Go Next