{"course":{"slug":"ai-images","title":"Image Generation, in Practice","url":"https://addaly.com/learn/ai-images","level":"beginner","track":"create"},"lesson":{"slug":"choosing-a-generator","title":"The generator you pick is a licensing decision","url":"https://addaly.com/learn/ai-images/choosing-a-generator","position":1,"of":69,"minutes":9,"answer":"The licence on a model's weights governs whether you may use it commercially, and it is a completely separate question from who owns the picture it made.","keyIdea":"The licence on a model's weights governs whether you may use it commercially, and it is a completely separate question from who owns the picture it made."},"body":{"format":"markdown","text":"## Everyone compares the pictures. Almost nobody compares the terms\n\nPick any two image generators and the sample galleries will look roughly as good as each other. That is not where they differ any more. They differ on three things that decide whether you can actually finish a job: what one image costs you, how much control you get when the first result is wrong, and what you are permitted to do with the file afterwards.\n\nThe third one is the one that ends careers-in-miniature: a designer who delivers 200 assets built on weights they were not licensed to use commercially.\n\nIf you have not read [how these models actually work](/learn/making-things-with-ai), read that first. This course assumes it and never repeats it.\n\n## The hosted tools, and what each is actually for\n\n**Midjourney** still produces the most immediately attractive frame with the least effort. The cost of that is an opinion baked into every image: a house look you have to fight. It is subscription-only, with no free tier, and on the entry plan your generations are visible in the public gallery; private generation sits on the higher tiers. If you work under NDA, check that before your first render, not after.\n\n**Adobe Firefly** is the conservative choice. It is trained on Adobe Stock and licensed or public-domain material, and Adobe offers indemnity against training-data claims on its business plans. It is less exciting and more defensible. It has a free monthly credit allowance, and Generative Fill inside Photoshop is the same engine.\n\n**Google's Gemini and Imagen line** is strong at following a long instruction and at editing an image you supply. Google stamps its generated images with SynthID, an invisible watermark. There is meaningful free use through the Gemini app and AI Studio.\n\n**Ideogram** exists because text inside images was broken for years. It renders legible words far better than the general-purpose models. It is still not typesetting, more on that in the finishing lesson.\n\n**Recraft** aims at design output rather than pictures: it will give you actual vector SVG, which matters if the thing you are making has to scale to a billboard. See [vector and print](/learn/vector-and-print) for why that distinction is not cosmetic.\n\n**Higgsfield** is a different species: a recipe-driven platform rather than a raw model. The next lesson is about that whole category.\n\n## Running the weights yourself\n\nThe open-weights line runs Stable Diffusion → SDXL → SD 3.x, alongside FLUX from Black Forest Labs and Qwen-Image from Alibaba. Running them yourself costs nothing per image, gives you every control that exists, and works offline.\n\nThe requirement is a GPU. SDXL is comfortable on 8 GB of VRAM. FLUX's larger variants want much more, though quantised versions run on 8-12 GB slowly. If you do not own a card: Google Colab's free tier, Kaggle's free weekly GPU hours, and Hugging Face Spaces all run these models at no cost. [Running models yourself](/learn/running-models-yourself) and [Hugging Face](/learn/hugging-face) cover the mechanics.\n\nFree front ends, all free: **ComfyUI** (a node graph, ugly, total control), **Fooocus** (Midjourney-like defaults, almost no settings), **InvokeAI**, and **Krita with the AI Diffusion plugin**, which is the best free inpainting experience there is.\n\n## The licence trap, stated plainly\n\nA model's weights carry a licence, and it is a separate thing from what you may do with the pictures.\n\n- **FLUX.1 [schnell]** is Apache 2.0. Do what you like.\n- **FLUX.1 [dev]** is a non-commercial licence. Thousands of people have downloaded it, run it locally, and billed clients for the output. That is a breach of the model licence, regardless of any question about who owns the image.\n- **Qwen-Image** was released Apache 2.0.\n- **Stability's recent releases** use a community licence with a revenue threshold: free below it, paid above.\n- **Hosted tools** replace all this with their terms of service, and free tiers frequently restrict commercial use even when paid tiers do not.\n\nThose were the positions in 2025 and they move. Check the model card on the day you start the job. Lesson ten deals with output rights, which is a different question again.\n\n## Choosing, in about thirty seconds\n\n- A single striking image, no budget for fiddling: Midjourney or the Gemini app.\n- Corporate work where somebody will ask about training data: Firefly, or Getty's and Shutterstock's own generators.\n- Words inside the image: Ideogram, then set real type over it anyway.\n- Hundreds of near-identical variants: a preset platform.\n- Precise control, a repeating character, or no money at all: open weights, locally or on free GPU hours.\n\nToday: open the model card or terms page of whatever you are currently using and find the sentence about commercial use. Most people have never read it."},"check":{"question":"A freelancer downloads FLUX.1 [dev], runs it on their own machine with no internet connection, and invoices a client for the resulting images. What is the actual problem?","options":[{"text":"Nothing is wrong. A licence attached to model weights controls redistributing the file, not what you make with it once it is on your disk.","correct":false,"tempting":"Reads a model licence as governing distribution only, when it also governs what the outputs may be used for."},{"text":"The images are unlikely to attract copyright, so there was nothing there to sell in the first place.","correct":false,"tempting":"Confuses whether the output is protectable with whether the freelancer was permitted to run the model at all."},{"text":"The [dev] weights are released under a non-commercial licence, so billing a client breaches the model's terms whatever the copyright position on the images turns out to be.","correct":true},{"text":"Generating locally strips the Content Credentials the client needed for disclosure, which is what makes the delivery non-compliant.","correct":false,"tempting":"Answers a provenance question when the question was about permission to use the model."}]},"prev":null,"next":{"slug":"presets-and-recipes","title":"A preset is somebody else's workflow, frozen","url":"https://addaly.com/learn/ai-images/presets-and-recipes"},"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) 'The generator you pick is a licensing decision', in Image Generation, in Practice. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-images/choosing-a-generator (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"The generator you pick is a licensing decision","courseTitle":"Image Generation, in Practice","module":{"index":1,"slug":"the-kit","title":"The kit, and what each piece of it costs you"},"answer":null,"keyIdea":"The licence on a model’s weights governs whether you may use it commercially, and it is a completely separate question from who owns the picture it made.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"everyone-compares-the-pictures-almost-nobody-compares-the-te","runs":[{"text":"Everyone compares the pictures. Almost nobody compares the terms"}]},{"t":"p","runs":[{"text":"Pick any two image generators and the sample galleries will look roughly as good as each other. That is not where they differ any more. They differ on three things that decide whether you can actually finish a job: what one image costs you, how much control you get when the first result is wrong, and what you are permitted to do with the file afterwards."}]},{"t":"p","runs":[{"text":"The third one is the one that ends careers-in-miniature: a designer who delivers 200 assets built on weights they were not licensed to use commercially."}]},{"t":"p","runs":[{"text":"If you have not read "},{"text":"how these models actually work","href":"https://addaly.com/learn/making-things-with-ai"},{"text":", read that first. This course assumes it and never repeats it."}]},{"t":"h2","id":"the-hosted-tools-and-what-each-is-actually-for","runs":[{"text":"The hosted tools, and what each is actually for"}]},{"t":"p","runs":[{"text":"Midjourney","b":true},{"text":" still produces the most immediately attractive frame with the least effort. The cost of that is an opinion baked into every image: a house look you have to fight. It is subscription-only, with no free tier, and on the entry plan your generations are visible in the public gallery; private generation sits on the higher tiers. If you work under NDA, check that before your first render, not after."}]},{"t":"p","runs":[{"text":"Adobe Firefly","b":true},{"text":" is the conservative choice. It is trained on Adobe Stock and licensed or public-domain material, and Adobe offers indemnity against training-data claims on its business plans. It is less exciting and more defensible. It has a free monthly credit allowance, and Generative Fill inside Photoshop is the same engine."}]},{"t":"p","runs":[{"text":"Google’s Gemini and Imagen line","b":true},{"text":" is strong at following a long instruction and at editing an image you supply. Google stamps its generated images with SynthID, an invisible watermark. There is meaningful free use through the Gemini app and AI Studio."}]},{"t":"p","runs":[{"text":"Ideogram","b":true},{"text":" exists because text inside images was broken for years. It renders legible words far better than the general-purpose models. It is still not typesetting, more on that in the finishing lesson."}]},{"t":"p","runs":[{"text":"Recraft","b":true},{"text":" aims at design output rather than pictures: it will give you actual vector SVG, which matters if the thing you are making has to scale to a billboard. See "},{"text":"vector and print","href":"https://addaly.com/learn/vector-and-print"},{"text":" for why that distinction is not cosmetic."}]},{"t":"p","runs":[{"text":"Higgsfield","b":true},{"text":" is a different species: a recipe-driven platform rather than a raw model. The next lesson is about that whole category."}]},{"t":"h2","id":"running-the-weights-yourself","runs":[{"text":"Running the weights yourself"}]},{"t":"p","runs":[{"text":"The open-weights line runs Stable Diffusion → SDXL → SD 3.x, alongside FLUX from Black Forest Labs and Qwen-Image from Alibaba. Running them yourself costs nothing per image, gives you every control that exists, and works offline."}]},{"t":"p","runs":[{"text":"The requirement is a GPU. SDXL is comfortable on 8 GB of VRAM. FLUX’s larger variants want much more, though quantised versions run on 8-12 GB slowly. If you do not own a card: Google Colab’s free tier, Kaggle’s free weekly GPU hours, and Hugging Face Spaces all run these models at no cost. "},{"text":"Running models yourself","href":"https://addaly.com/learn/running-models-yourself"},{"text":" and "},{"text":"Hugging Face","href":"https://addaly.com/learn/hugging-face"},{"text":" cover the mechanics."}]},{"t":"p","runs":[{"text":"Free front ends, all free: "},{"text":"ComfyUI","b":true},{"text":" (a node graph, ugly, total control), "},{"text":"Fooocus","b":true},{"text":" (Midjourney-like defaults, almost no settings), "},{"text":"InvokeAI","b":true},{"text":", and "},{"text":"Krita with the AI Diffusion plugin","b":true},{"text":", which is the best free inpainting experience there is."}]},{"t":"h2","id":"the-licence-trap-stated-plainly","runs":[{"text":"The licence trap, stated plainly"}]},{"t":"p","runs":[{"text":"A model’s weights carry a licence, and it is a separate thing from what you may do with the pictures."}]},{"t":"ul","items":[[{"text":"FLUX.1 [schnell]","b":true},{"text":" is Apache 2.0. Do what you like."}],[{"text":"FLUX.1 [dev]","b":true},{"text":" is a non-commercial licence. Thousands of people have downloaded it, run it locally, and billed clients for the output. That is a breach of the model licence, regardless of any question about who owns the image."}],[{"text":"Qwen-Image","b":true},{"text":" was released Apache 2.0."}],[{"text":"Stability’s recent releases","b":true},{"text":" use a community licence with a revenue threshold: free below it, paid above."}],[{"text":"Hosted tools","b":true},{"text":" replace all this with their terms of service, and free tiers frequently restrict commercial use even when paid tiers do not."}]]},{"t":"p","runs":[{"text":"Those were the positions in 2025 and they move. Check the model card on the day you start the job. Lesson ten deals with output rights, which is a different question again."}]},{"t":"h2","id":"choosing-in-about-thirty-seconds","runs":[{"text":"Choosing, in about thirty seconds"}]},{"t":"ul","items":[[{"text":"A single striking image, no budget for fiddling: Midjourney or the Gemini app."}],[{"text":"Corporate work where somebody will ask about training data: Firefly, or Getty’s and Shutterstock’s own generators."}],[{"text":"Words inside the image: Ideogram, then set real type over it anyway."}],[{"text":"Hundreds of near-identical variants: a preset platform."}],[{"text":"Precise control, a repeating character, or no money at all: open weights, locally or on free GPU hours."}]]},{"t":"p","runs":[{"text":"Today: open the model card or terms page of whatever you are currently using and find the sentence about commercial use. Most people have never read it."}]}],"check":{"question":[{"text":"A freelancer downloads FLUX.1 [dev], runs it on their own machine with no internet connection, and invoices a client for the resulting images. What is the actual problem?"}],"options":[{"text":[{"text":"Nothing is wrong. A licence attached to model weights controls redistributing the file, not what you make with it once it is on your disk."}],"correct":false,"tempting":[{"text":"Reads a model licence as governing distribution only, when it also governs what the outputs may be used for."}]},{"text":[{"text":"The images are unlikely to attract copyright, so there was nothing there to sell in the first place."}],"correct":false,"tempting":[{"text":"Confuses whether the output is protectable with whether the freelancer was permitted to run the model at all."}]},{"text":[{"text":"The [dev] weights are released under a non-commercial licence, so billing a client breaches the model’s terms whatever the copyright position on the images turns out to be."}],"correct":true,"tempting":null},{"text":[{"text":"Generating locally strips the Content Credentials the client needed for disclosure, which is what makes the delivery non-compliant."}],"correct":false,"tempting":[{"text":"Answers a provenance question when the question was about permission to use the model."}]}],"order":[1,3,2,0],"why":null},"prev":null,"next":{"slug":"presets-and-recipes","title":"A preset is somebody else’s workflow, frozen","minutes":8}}}