I'm trying to draw more realistic portraits but I get stuck looking at AI-generated images from Midjourney and DALL-E. Is using those images as reference the same as copying? Is there an ethical way to use these tools to study proportions and shadows without feeling like I'm cheating?
It's not really copying, but I get that feeling! The AI-generated image is just a visual starting point - what matters is observing *how* the AI solved the proportions, where it placed the shadows, how it handled the planes of light - and then drawing that with your own hand, based on your understanding. If you're screenshotting Midjourney and drawing over it pixel by pixel, then yeah, that's copying; but if you use it to study "oh, I see the apple's shadow drops at a 45-degree angle and gets darker near the center" and then draw a different object applying that knowledge, you're actually learning! The ethical question resolves itself when your final work is completely yours - neither the subject, nor the composition, nor the style need to come from the reference image you used.
The question is really about *intentionality* rather than the tool itself - when I see a reference, whether it's AI-generated or a photograph, what interests me is breaking down the visual problem that's right in front of me. If you use the generated image just to understand "oh, the shadow starts at the chin and goes down like this," and then you go draw based on that observation, you're doing legitimate study. The real risk is when the generated image looks so good that you get tempted to just trace the outlines - that's when you lose the point.
My advice is this: generate as many images as you want, but then step away from the screen and draw from memory what you observed. If you can reproduce the proportions and light treatment without looking at it, that means you've actually learned something. And honestly, AI images often have little quirks in certain areas - fingers, ears, reflections - that end up being great points for critical analysis. You're not fooling anyone, you're studying.
A lot of people don't realize that AI-generated images tend to make the same mistakes over and over - weird hands, facial proportions that don't add up, anatomically nonsensical details. That's actually a huge *advantage* for anyone who wants to study seriously, because it forces you to think critically about what you're seeing instead of just copying mechanically.
The best way to use this ethically is to treat it as an anatomical and lighting reference, not as a "ready-made recipe". Generate the image, then study it: why did the face end up proportioned or not? How does the shadow define volume? Take screenshots of specific details (an eye, a hand, a facial plane) and work on those isolated elements, mixing them with real photographic studies. That way you're building genuine visual understanding, not xeroxing. The ethical difference is in not signing off your drawing as if it were 100% your own creation when you know it traced directly from it - but learning through that image? That's legit.
Another thing: take advantage of the mistakes themselves. When you notice something off in an AI generation, *draw the correct version*. That's real anatomy practice. And alternate with photography, because the camera captures nuances that AI still fails at - skin textures, unique characteristics that no generation can replicate. In the end, mixing it all together (AI, photo, live observation) is what develops your eye. 🎨
AI references work best when you use them to *disagree* with them - literally, write down the errors you find (those weird hands, proportions that don't add up) and redraw them corrected. That forces your brain to actually understand anatomy principles instead of just copying what you see. Also try generating several versions of the same face and comparing how the AI interpreted different angles, because that way you're studying *variation* within the same features - it's like having a model that shifts slightly each time, which is great for learning flexibility.
Log into answer.
