{"course":{"slug":"ai-images","title":"Image Generation, in Practice","url":"https://addaly.com/learn/ai-images","level":"beginner","track":"create"},"lesson":{"slug":"presets-and-recipes","title":"A preset is somebody else's workflow, frozen","url":"https://addaly.com/learn/ai-images/presets-and-recipes","position":2,"of":69,"minutes":8,"answer":"Recipe-driven tools trade control for speed, and the bill arrives at the first revision, when the thing the client wants changed is the thing the preset already decided.","keyIdea":"Recipe-driven tools trade control for speed, and the bill arrives at the first revision, when the thing the client wants changed is the thing the preset already decided."},"body":{"format":"markdown","text":"## Why the packaged tools feel like magic\n\nOpen a raw model and you face a blank prompt box, a sampler dropdown, a step count, a guidance scale, an aspect ratio and a seed. Most of those decisions have one sensible answer for the thing you are making, and you do not know what it is yet.\n\nOpen a platform like **Higgsfield** and you face a wall of named recipes instead: a product-shot look, a marketing card layout, a cinematic portrait, a set of camera moves borrowed from its video side. You click one, you drop in a photo or a line of text, and something usable comes out in under a minute.\n\nThat is not a cheap trick. A preset is a real workflow that somebody built and tested: a chosen model, a prompt scaffold you never see, probably a style adapter or a LoRA, a fixed aspect ratio, a lighting description, sometimes an upscale pass on the end. The speed is genuine. So is the price you pay for it.\n\n## What Higgsfield-shaped tools are good at\n\nThree jobs, and they are good at all three.\n\n**Volume with a consistent look.** Forty product cards a week that must belong to the same family. A raw model gives you forty siblings that do not match. A preset gives you forty that do, because the same hidden scaffold is under each one.\n\n**Character identity as a saved object.** The interesting feature in this category is identity training: Higgsfield calls its version Soul ID. You supply a set of photographs of a person once, the platform trains an identity, and then that person can appear across many generations without you rebuilding the setup each time. Under the surface this is the same idea as a LoRA, which lesson six covers properly; the platform difference is that it is three clicks instead of a training script.\n\n**Camera language as a control.** Because tools like this grew out of video, they expose motion and lens vocabulary as buttons (crash zoom, dolly, low angle, bullet time) rather than expecting you to write it into a prompt and hope. If you do not have a cinematography vocabulary yet, this teaches you one by letting you see each term applied to your own frame.\n\n## What you give up, and when you notice\n\nYou notice at the revision.\n\nThe client likes the bottle shot but wants the light coming from behind the glass instead of the front. That decision was made inside the preset. It is not in your prompt, so no amount of rewriting the prompt reaches it. The preset's opinion about lighting is applied after your words and overrides them. You cannot get to a point halfway between two presets, either: they are discrete choices, not sliders.\n\nThis is the general rule and it is worth carrying beyond this one tool: **anything that packages a workflow buys you speed by removing exactly the controls you will need when the first answer is wrong.** It is a good trade for the ninety per cent of work that is routine, and a bad trade for the ten per cent that is the reason a client hired a person.\n\n## The credit arithmetic nobody checks\n\nPlatforms in this category run on credits, and within one platform the per-image credit cost across the available models can differ by nearly an order of magnitude.\n\nA concrete measurement, from a real project in mid-2026: generating small avatar portraits on one such platform, a lightweight model cost 0.15 credits per image while two premium models cost 1.00 and 1.25. Seven to eight times the price. At the size the images were actually used (a 48-pixel avatar) no one could tell them apart.\n\nThe lesson generalises. Before you commit a batch, generate the same frame on the cheapest and the dearest model the platform offers, then look at both **at the size they will be seen**. Half the time the expensive one wins clearly and is worth it. The other half you have just found a way to make your allowance last eight times longer.\n\n## The free path\n\nThere is no free tier that gives you a preset platform. What there is, and what does the same job: **Fooocus**, which is a free local front end built on exactly this philosophy (good defaults, almost no visible settings, style presets in a dropdown) and runs on free Colab or Kaggle GPU hours if you have no card. **Canva's free tier** packages generation into templates in a comparable way for social and print layouts, and works on a phone. Neither will give you trained character identity as a one-click object; you build that yourself in lesson six.\n\n## Use one honestly\n\nA preset tool is not a lesser tool. It is a tool with a shape. Use it when the work is repetitive and the look is agreed. Reach for raw model controls when a specific person has a specific note.\n\nToday: pick any preset you like the output of, then try to produce the *same* image with one deliberate change: a different light direction, a different lens. How far you get tells you exactly how much control that platform kept for itself."},"check":{"question":"A team produces 40 product cards a week on a preset platform with no complaints. Then a client asks for one card where the light comes from behind the bottle rather than the front, and nobody on the team can produce it. What does this actually reveal?","options":[{"text":"The underlying model is too small to follow lighting instructions, so the team needs a platform running a larger model.","correct":false,"tempting":"Blames model capacity for a control the tool never exposed in the first place."},{"text":"The preset has already fixed the lighting decision and applies it regardless of the prompt, so the fix is a tool that exposes lighting directly, or a manual relight in a raster editor.","correct":true},{"text":"The prompt needs to name the backlighting explicitly and in more detail than it currently does.","correct":false,"tempting":"Assumes prompt wording can override conditioning that the preset applies after the prompt."},{"text":"The preset needs retraining on backlit examples before it can produce them.","correct":false,"tempting":"Treats a packaged workflow as a model the user can retrain, rather than a fixed set of settings."}]},"prev":{"slug":"choosing-a-generator","title":"The generator you pick is a licensing decision","url":"https://addaly.com/learn/ai-images/choosing-a-generator"},"next":{"slug":"hosted-or-local","title":"Three routes to a GPU, and the arithmetic for each","url":"https://addaly.com/learn/ai-images/hosted-or-local"},"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) 'A preset is somebody else's workflow, frozen', in Image Generation, in Practice. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-images/presets-and-recipes (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":"A preset is somebody else’s workflow, frozen","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":"Recipe-driven tools trade control for speed, and the bill arrives at the first revision, when the thing the client wants changed is the thing the preset already decided.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"why-the-packaged-tools-feel-like-magic","runs":[{"text":"Why the packaged tools feel like magic"}]},{"t":"p","runs":[{"text":"Open a raw model and you face a blank prompt box, a sampler dropdown, a step count, a guidance scale, an aspect ratio and a seed. Most of those decisions have one sensible answer for the thing you are making, and you do not know what it is yet."}]},{"t":"p","runs":[{"text":"Open a platform like "},{"text":"Higgsfield","b":true},{"text":" and you face a wall of named recipes instead: a product-shot look, a marketing card layout, a cinematic portrait, a set of camera moves borrowed from its video side. You click one, you drop in a photo or a line of text, and something usable comes out in under a minute."}]},{"t":"p","runs":[{"text":"That is not a cheap trick. A preset is a real workflow that somebody built and tested: a chosen model, a prompt scaffold you never see, probably a style adapter or a LoRA, a fixed aspect ratio, a lighting description, sometimes an upscale pass on the end. The speed is genuine. So is the price you pay for it."}]},{"t":"h2","id":"what-higgsfield-shaped-tools-are-good-at","runs":[{"text":"What Higgsfield-shaped tools are good at"}]},{"t":"p","runs":[{"text":"Three jobs, and they are good at all three."}]},{"t":"p","runs":[{"text":"Volume with a consistent look.","b":true},{"text":" Forty product cards a week that must belong to the same family. A raw model gives you forty siblings that do not match. A preset gives you forty that do, because the same hidden scaffold is under each one."}]},{"t":"p","runs":[{"text":"Character identity as a saved object.","b":true},{"text":" The interesting feature in this category is identity training: Higgsfield calls its version Soul ID. You supply a set of photographs of a person once, the platform trains an identity, and then that person can appear across many generations without you rebuilding the setup each time. Under the surface this is the same idea as a LoRA, which lesson six covers properly; the platform difference is that it is three clicks instead of a training script."}]},{"t":"p","runs":[{"text":"Camera language as a control.","b":true},{"text":" Because tools like this grew out of video, they expose motion and lens vocabulary as buttons (crash zoom, dolly, low angle, bullet time) rather than expecting you to write it into a prompt and hope. If you do not have a cinematography vocabulary yet, this teaches you one by letting you see each term applied to your own frame."}]},{"t":"h2","id":"what-you-give-up-and-when-you-notice","runs":[{"text":"What you give up, and when you notice"}]},{"t":"p","runs":[{"text":"You notice at the revision."}]},{"t":"p","runs":[{"text":"The client likes the bottle shot but wants the light coming from behind the glass instead of the front. That decision was made inside the preset. It is not in your prompt, so no amount of rewriting the prompt reaches it. The preset’s opinion about lighting is applied after your words and overrides them. You cannot get to a point halfway between two presets, either: they are discrete choices, not sliders."}]},{"t":"p","runs":[{"text":"This is the general rule and it is worth carrying beyond this one tool: "},{"text":"anything that packages a workflow buys you speed by removing exactly the controls you will need when the first answer is wrong.","b":true},{"text":" It is a good trade for the ninety per cent of work that is routine, and a bad trade for the ten per cent that is the reason a client hired a person."}]},{"t":"h2","id":"the-credit-arithmetic-nobody-checks","runs":[{"text":"The credit arithmetic nobody checks"}]},{"t":"p","runs":[{"text":"Platforms in this category run on credits, and within one platform the per-image credit cost across the available models can differ by nearly an order of magnitude."}]},{"t":"p","runs":[{"text":"A concrete measurement, from a real project in mid-2026: generating small avatar portraits on one such platform, a lightweight model cost 0.15 credits per image while two premium models cost 1.00 and 1.25. Seven to eight times the price. At the size the images were actually used (a 48-pixel avatar) no one could tell them apart."}]},{"t":"p","runs":[{"text":"The lesson generalises. Before you commit a batch, generate the same frame on the cheapest and the dearest model the platform offers, then look at both "},{"text":"at the size they will be seen","b":true},{"text":". Half the time the expensive one wins clearly and is worth it. The other half you have just found a way to make your allowance last eight times longer."}]},{"t":"h2","id":"the-free-path","runs":[{"text":"The free path"}],"role":"free"},{"t":"p","runs":[{"text":"There is no free tier that gives you a preset platform. What there is, and what does the same job: "},{"text":"Fooocus","b":true},{"text":", which is a free local front end built on exactly this philosophy (good defaults, almost no visible settings, style presets in a dropdown) and runs on free Colab or Kaggle GPU hours if you have no card. "},{"text":"Canva’s free tier","b":true},{"text":" packages generation into templates in a comparable way for social and print layouts, and works on a phone. Neither will give you trained character identity as a one-click object; you build that yourself in lesson six."}]},{"t":"h2","id":"use-one-honestly","runs":[{"text":"Use one honestly"}]},{"t":"p","runs":[{"text":"A preset tool is not a lesser tool. It is a tool with a shape. Use it when the work is repetitive and the look is agreed. Reach for raw model controls when a specific person has a specific note."}]},{"t":"p","runs":[{"text":"Today: pick any preset you like the output of, then try to produce the "},{"text":"same","i":true},{"text":" image with one deliberate change: a different light direction, a different lens. How far you get tells you exactly how much control that platform kept for itself."}]}],"check":{"question":[{"text":"A team produces 40 product cards a week on a preset platform with no complaints. Then a client asks for one card where the light comes from behind the bottle rather than the front, and nobody on the team can produce it. What does this actually reveal?"}],"options":[{"text":[{"text":"The underlying model is too small to follow lighting instructions, so the team needs a platform running a larger model."}],"correct":false,"tempting":[{"text":"Blames model capacity for a control the tool never exposed in the first place."}]},{"text":[{"text":"The preset has already fixed the lighting decision and applies it regardless of the prompt, so the fix is a tool that exposes lighting directly, or a manual relight in a raster editor."}],"correct":true,"tempting":null},{"text":[{"text":"The prompt needs to name the backlighting explicitly and in more detail than it currently does."}],"correct":false,"tempting":[{"text":"Assumes prompt wording can override conditioning that the preset applies after the prompt."}]},{"text":[{"text":"The preset needs retraining on backlit examples before it can produce them."}],"correct":false,"tempting":[{"text":"Treats a packaged workflow as a model the user can retrain, rather than a fixed set of settings."}]}],"order":[0,1,3,2],"why":null},"prev":{"slug":"choosing-a-generator","title":"The generator you pick is a licensing decision","minutes":9},"next":{"slug":"hosted-or-local","title":"Three routes to a GPU, and the arithmetic for each","minutes":9}}}