--- title: Jacob Miller - DiffUI | daily.dev description: A podcast interview with Jacob Miller, former Figma design systems lead, about building DiffUI, a diffusion-model-based UI design tool. He explains why he bet... canonical: https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Jacob Miller - DiffUI | daily.dev og:description: A podcast interview with Jacob Miller, former Figma design systems lead, about building DiffUI, a diffusion-model-based UI design tool. He explains why he bet... og:url: https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0 og:image: https://api.daily.dev/og/posts/BzHGiwFr0.png og:image:alt: Jacob Miller - DiffUI og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Jacob Miller - DiffUI **[devtools-fm](https://daily.dev/sources/devtoolsfm)** · 53 min read · 0 upvotes · 0 comments ## Summary A podcast interview with Jacob Miller, former Figma design systems lead, about building DiffUI, a diffusion-model-based UI design tool. He explains why he bet on diffusion models over LLM-generated code for design, how GPT image 2 became a major unlock, why roughly 60% of his users turned out to be engineers rather than designers, and how the product uses a node-based canvas with a semantic JSON layer to maintain brand and design-token consistency across generated pages. He also discusses usage-based pricing (13 cents per generated image), argues traditional rigid design systems are becoming obsolete due to AI, and predicts product manager, designer, and engineer roles will converge into generalists. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why did Figma choose LLM-based code generation over diffusion models for its prompt-to-design feature? Figma went with large language models generating code because it was the safer, more structured option with a clear path to victory, especially with a major conference deadline five months away. Diffusion models at the time only targeted consumer-level GPUs and hadn't been scaled on data center hardware the way LLMs had, and they also struggled with rendering text and fine details like icons. _Engineers weighing LLM versus diffusion approaches for design tooling can track these tradeoffs on daily.dev._ ### How much does it cost to generate a full website design using a diffusion model based tool instead of an LLM code generator like Claude? DiffUI charges 13 cents per generated image at zero-margin API pricing, and a full multi-page site design typically takes around 30 images, costing under $4 total. This is presented as far cheaper than LLM-based tools like Claude design, because diffusion models avoid loading the entire design system and codebase into the context window for every generation. _daily.dev helps developers comparing AI design tool pricing keep track of usage-based cost models._ ### How do AI design tools avoid producing generic looking sites with the same visual tells like eyebrows and stars? Diffusion models trained on real web screenshots tend to avoid the repetitive AI-generated visual patterns like eyebrows, serif accents, and stars that emerge when large language models write consistent code, because diffusion operates at higher temperature and broader variation than code-generating LLMs. However, this advantage may erode over time as diffusion models get trained on screenshots of other AI-generated sites, gravitating toward the same slop. _daily.dev keeps developers building with generative design tools aware of emerging AI design pattern pitfalls._ ## Similar posts on daily.dev - [Medium](https://daily.dev/posts/medium-yte4khgst) · Medium · 1 upvotes · 0 comments - [What the design-to-code loop unlocks](https://daily.dev/posts/what-the-design-to-code-loop-unlocks-rutzwo1xa) · Figma · 0 upvotes · 0 comments - [Generative UI Notes](https://daily.dev/posts/generative-ui-notes-2xfiiqqj9) · CSS-Tricks · 5 upvotes · 0 comments --- Tags: [#figma](https://daily.dev/tags/figma), [#design-systems](https://daily.dev/tags/design-systems), [#diffusion-models](https://daily.dev/tags/diffusion-models) [View this post on daily.dev](https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Jacob Miller - DiffUI","url":"https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0"},"datePublished":"2026-08-17T06:47:47.838Z","dateModified":"2026-08-17T08:35:04.958Z","description":"A podcast interview with Jacob Miller, former Figma design systems lead, about building DiffUI, a diffusion-model-based UI design tool. He explains why he bet...","image":"https://i.ytimg.com/vi/wu00N4PWBFk/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/wu00N4PWBFk/sddefault.jpg","isAccessibleForFree":true,"articleSection":"devtools-fm","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"devtools-fm","logo":"https://media.daily.dev/image/upload/s--yvUzE7NR--/f_auto/v1717745371/logos/devtoolsfm","url":"https://daily.dev/sources/devtoolsfm"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"figma,design-systems,diffusion-models","timeRequired":"PT53M","video":{"@type":"VideoObject","name":"Jacob Miller - DiffUI","description":"A podcast interview with Jacob Miller, former Figma design systems lead, about building DiffUI, a diffusion-model-based UI design tool. He explains why he bet...","thumbnailUrl":"https://i.ytimg.com/vi/wu00N4PWBFk/sddefault.jpg","uploadDate":"2026-08-17T06:47:47.838Z","duration":"PT53M","url":"https://api.daily.dev/r/BzHGiwFr0","embedUrl":"https://www.youtube.com/embed/wu00N4PWBFk"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"devtools-fm","item":"https://daily.dev/sources/devtoolsfm"},{"@type":"ListItem","position":3,"name":"Jacob Miller - DiffUI"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/jacob-miller---diffui-bzhgiwfr0#faq","mainEntity":[{"@type":"Question","name":"Why did Figma choose LLM-based code generation over diffusion models for its prompt-to-design feature?","acceptedAnswer":{"@type":"Answer","text":"Figma went with large language models generating code because it was the safer, more structured option with a clear path to victory, especially with a major conference deadline five months away. Diffusion models at the time only targeted consumer-level GPUs and hadn't been scaled on data center hardware the way LLMs had, and they also struggled with rendering text and fine details like icons. Engineers weighing LLM versus diffusion approaches for design tooling can track these tradeoffs on daily.dev."}},{"@type":"Question","name":"How much does it cost to generate a full website design using a diffusion model based tool instead of an LLM code generator like Claude?","acceptedAnswer":{"@type":"Answer","text":"DiffUI charges 13 cents per generated image at zero-margin API pricing, and a full multi-page site design typically takes around 30 images, costing under $4 total. This is presented as far cheaper than LLM-based tools like Claude design, because diffusion models avoid loading the entire design system and codebase into the context window for every generation. daily.dev helps developers comparing AI design tool pricing keep track of usage-based cost models."}},{"@type":"Question","name":"How do AI design tools avoid producing generic looking sites with the same visual tells like eyebrows and stars?","acceptedAnswer":{"@type":"Answer","text":"Diffusion models trained on real web screenshots tend to avoid the repetitive AI-generated visual patterns like eyebrows, serif accents, and stars that emerge when large language models write consistent code, because diffusion operates at higher temperature and broader variation than code-generating LLMs. However, this advantage may erode over time as diffusion models get trained on screenshots of other AI-generated sites, gravitating toward the same slop. daily.dev keeps developers building with generative design tools aware of emerging AI design pattern pitfalls."}}]} ```