{"course":{"slug":"ai-actually-explained","title":"AI, Actually Explained","url":"https://addaly.com/learn/ai-actually-explained","level":"beginner","track":"start"},"lesson":{"slug":"how-the-prompt-reaches-the-picture","title":"How words steer the picture","url":"https://addaly.com/learn/ai-actually-explained/how-the-prompt-reaches-the-picture","position":54,"of":92,"minutes":9,"answer":"The prompt is turned into a bag of concept-positions by a separate model, which is why an image prompt behaves like a list of nouns, why \"no elephants\" draws elephants.","keyIdea":"The prompt is turned into a bag of concept-positions by a separate model, which is why an image prompt behaves like a list of nouns, why \"no elephants\" draws elephants, and why a stronger text reader fixed most of it."},"body":{"format":"markdown","text":"## Two models, not one\n\nWhen you type a prompt into an image generator, the diffusion network you met in the previous lesson never sees your words. A separate model reads them first and converts them into a set of numbers: the same idea as the embeddings in the chatbot module, where meaning became a position in space. Those numbers are handed to the denoiser at every step, and the denoiser steers the emerging image toward them.\n\nEverything strange about image prompts follows from what that reader is, and what it was trained to do.\n\n## The reader that learned from captions\n\nThe reader used by the first public models was trained in 2021 on four hundred million pairs of images and their captions, scraped from the web. Its job was simple: given an image and a caption, place them near each other in the space; given an image and someone else's caption, place them far apart. It learned which words go with which pictures.\n\nIt did not learn grammar, because captions do not need it. \"A dog chasing a cat\" and \"a cat chasing a dog\" have the same words and, in this reader, nearly the same position. It did not learn negation, because almost no caption says what is *not* in the picture. And it did not learn letters, because it saw words as whole tokens and never needed to know what a \"B\" looks like.\n\n## Three consequences you will meet immediately\n\n**Prompts behave like a list.** Since the reader keeps concepts and drops structure, a prompt that is a comma-separated list of nouns and styles (\"portrait, elderly fisherman, harbour at dawn, film grain\") works about as well as a sentence, and often better. The folk wisdom is right; the mechanism is a reader that cannot use the sentence anyway.\n\n**\"No elephants\" draws elephants.** The word \"elephant\" enters the space as the concept of an elephant; the word \"no\" has almost no pull. The picture is steered toward elephants. Image tools added a separate box (a negative prompt) for exactly this reason. Technically the process is run twice, once toward your prompt and once toward the negative one, and the second is subtracted from the first. That subtraction is also how the strength of the steering is set: a \"guidance\" number, typically around seven, that says how far to push toward the words. Set it low and you get a loose, varied picture that ignores you; set it high and you get an over-saturated one that obeys every noun too hard.\n\n**Order of words matters a little, and position in the image not at all.** \"A red cube on a blue sphere\" comes out as red and blue and a cube and a sphere in some arrangement. The reader has no left, right, above or below.\n\n## What fixed it\n\nFrom 2022 onward the labs replaced or supplemented the caption-trained reader with a language model of the kind in the chatbot module: one trained on text, with grammar, that knows what \"on\" and \"not\" mean. Google's 2022 model did this first; the third generation of Stable Diffusion and the Flux models did it in 2024. Two other changes worked alongside it. Training captions were rewritten by a vision model into long, detailed descriptions, because the web's own captions are short and vague, and a model trained on \"IMG_4032.jpg\" learns nothing from it. And the readers began to see letters, which is a large part of why text in images went from gibberish to usable.\n\nNone of this made the model understand the sentence. It made the reader's positions carry more of the sentence's structure than before. Ask a 2025 model for a red cube on a blue sphere and it will usually oblige. Ask for five objects in a specific arrangement with two of them upside down and you will discover where the improvement stops.\n\n## What this means in practice\n\nDescribe what you want to *see*, not what you want to avoid. Put the subject first. Name the medium and the light. If the tool has a negative box, use it for the things that keep appearing. And when a model refuses to do what a sentence says, it is not being difficult; the reader has flattened your sentence into a cloud of concepts, and you are asking the cloud to have a shape it cannot hold. Rephrase toward the list, or reach for an editing tool that lets you place things by hand: every free tool named in the previous lesson has one."},"check":{"question":"A prompt reading \"a beach with no people\" produces a crowded beach. What is the mechanism?","options":[{"text":"The model is trained to make images look lively, so it adds people to empty scenes.","correct":false,"tempting":"Invents a preference; the model has no notion of lively, only of which words pull toward which pictures."},{"text":"The word \"people\" enters the reader as a concept with pull, and \"no\" has almost none, so the image is steered toward people.","correct":true},{"text":"The model misread the prompt because nearly every beach in its training data contained people, so it could not draw an empty one.","correct":false,"tempting":"Blames the data for a failure that comes from the reader; the same prompt fails for objects rare on beaches."},{"text":"The prompt was too short for the model to understand the request.","correct":false,"tempting":"Assumes length is the problem; a longer prompt with the same negation fails the same way."}]},"prev":{"slug":"a-picture-from-noise","title":"A picture, out of noise","url":"https://addaly.com/learn/ai-actually-explained/a-picture-from-noise"},"next":{"slug":"why-the-hands-were-wrong","title":"Why the hands were wrong","url":"https://addaly.com/learn/ai-actually-explained/why-the-hands-were-wrong"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'How words steer the picture', in AI, Actually Explained. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-actually-explained/how-the-prompt-reaches-the-picture (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"How words steer the picture","courseTitle":"AI, Actually Explained","module":{"index":6,"slug":"beyond-text","title":"Beyond text: pictures, voices and video"},"answer":null,"keyIdea":"The prompt is turned into a bag of concept-positions by a separate model, which is why an image prompt behaves like a list of nouns, why “no elephants” draws elephants, and why a stronger text reader fixed most of it.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"two-models-not-one","runs":[{"text":"Two models, not one"}]},{"t":"p","runs":[{"text":"When you type a prompt into an image generator, the diffusion network you met in the previous lesson never sees your words. A separate model reads them first and converts them into a set of numbers: the same idea as the embeddings in the chatbot module, where meaning became a position in space. Those numbers are handed to the denoiser at every step, and the denoiser steers the emerging image toward them."}]},{"t":"p","runs":[{"text":"Everything strange about image prompts follows from what that reader is, and what it was trained to do."}]},{"t":"h2","id":"the-reader-that-learned-from-captions","runs":[{"text":"The reader that learned from captions"}]},{"t":"p","runs":[{"text":"The reader used by the first public models was trained in 2021 on four hundred million pairs of images and their captions, scraped from the web. Its job was simple: given an image and a caption, place them near each other in the space; given an image and someone else’s caption, place them far apart. It learned which words go with which pictures."}]},{"t":"p","runs":[{"text":"It did not learn grammar, because captions do not need it. “A dog chasing a cat” and “a cat chasing a dog” have the same words and, in this reader, nearly the same position. It did not learn negation, because almost no caption says what is "},{"text":"not","i":true},{"text":" in the picture. And it did not learn letters, because it saw words as whole tokens and never needed to know what a “B” looks like."}]},{"t":"h2","id":"three-consequences-you-will-meet-immediately","runs":[{"text":"Three consequences you will meet immediately"}]},{"t":"p","runs":[{"text":"Prompts behave like a list.","b":true},{"text":" Since the reader keeps concepts and drops structure, a prompt that is a comma-separated list of nouns and styles (“portrait, elderly fisherman, harbour at dawn, film grain”) works about as well as a sentence, and often better. The folk wisdom is right; the mechanism is a reader that cannot use the sentence anyway."}]},{"t":"p","runs":[{"text":"”No elephants” draws elephants.","b":true},{"text":" The word “elephant” enters the space as the concept of an elephant; the word “no” has almost no pull. The picture is steered toward elephants. Image tools added a separate box (a negative prompt) for exactly this reason. Technically the process is run twice, once toward your prompt and once toward the negative one, and the second is subtracted from the first. That subtraction is also how the strength of the steering is set: a “guidance” number, typically around seven, that says how far to push toward the words. Set it low and you get a loose, varied picture that ignores you; set it high and you get an over-saturated one that obeys every noun too hard."}]},{"t":"p","runs":[{"text":"Order of words matters a little, and position in the image not at all.","b":true},{"text":" “A red cube on a blue sphere” comes out as red and blue and a cube and a sphere in some arrangement. The reader has no left, right, above or below."}]},{"t":"h2","id":"what-fixed-it","runs":[{"text":"What fixed it"}]},{"t":"p","runs":[{"text":"From 2022 onward the labs replaced or supplemented the caption-trained reader with a language model of the kind in the chatbot module: one trained on text, with grammar, that knows what “on” and “not” mean. Google’s 2022 model did this first; the third generation of Stable Diffusion and the Flux models did it in 2024. Two other changes worked alongside it. Training captions were rewritten by a vision model into long, detailed descriptions, because the web’s own captions are short and vague, and a model trained on “IMG_4032.jpg” learns nothing from it. And the readers began to see letters, which is a large part of why text in images went from gibberish to usable."}]},{"t":"p","runs":[{"text":"None of this made the model understand the sentence. It made the reader’s positions carry more of the sentence’s structure than before. Ask a 2025 model for a red cube on a blue sphere and it will usually oblige. Ask for five objects in a specific arrangement with two of them upside down and you will discover where the improvement stops."}]},{"t":"h2","id":"what-this-means-in-practice","runs":[{"text":"What this means in practice"}]},{"t":"p","runs":[{"text":"Describe what you want to "},{"text":"see","i":true},{"text":", not what you want to avoid. Put the subject first. Name the medium and the light. If the tool has a negative box, use it for the things that keep appearing. And when a model refuses to do what a sentence says, it is not being difficult; the reader has flattened your sentence into a cloud of concepts, and you are asking the cloud to have a shape it cannot hold. Rephrase toward the list, or reach for an editing tool that lets you place things by hand: every free tool named in the previous lesson has one."}]}],"check":{"question":[{"text":"A prompt reading “a beach with no people” produces a crowded beach. What is the mechanism?"}],"options":[{"text":[{"text":"The model is trained to make images look lively, so it adds people to empty scenes."}],"correct":false,"tempting":[{"text":"Invents a preference; the model has no notion of lively, only of which words pull toward which pictures."}]},{"text":[{"text":"The word “people” enters the reader as a concept with pull, and “no” has almost none, so the image is steered toward people."}],"correct":true,"tempting":null},{"text":[{"text":"The model misread the prompt because nearly every beach in its training data contained people, so it could not draw an empty one."}],"correct":false,"tempting":[{"text":"Blames the data for a failure that comes from the reader; the same prompt fails for objects rare on beaches."}]},{"text":[{"text":"The prompt was too short for the model to understand the request."}],"correct":false,"tempting":[{"text":"Assumes length is the problem; a longer prompt with the same negation fails the same way."}]}],"order":[3,1,0,2],"why":null},"prev":{"slug":"a-picture-from-noise","title":"A picture, out of noise","minutes":9},"next":{"slug":"why-the-hands-were-wrong","title":"Why the hands were wrong","minutes":8}}}