AI Design Best of the Day · Monday, Nov 9, 2026
Claude Fable 5.2 lands and makers point it at Blender and 3D within hours, while Epic ships an Unreal Engine preview it says was built mostly by agents
The pick of the day is Anthropic's Claude Fable 5.2, released today, and the first wave of design tests that followed it. The launch page leans on design-adjacent claims, including three working electronic devices with 3D-printed enclosures and a hardware-design benchmark, and it was the top AI story on Hacker News this afternoon. Within three hours, @LexnLin's "took ONLY about 30minutes" clip had passed 45,000 views and makers were posting Blender builds, a sound-design test and a code-only jet. Around it: Epic released the Unreal Engine 5.9 MegaJam Preview, which puts Claude Code, Codex and Cursor inside the editor and adds an AI image, video and mesh tool called Unreal Canvas. ByteDance launched Seedream 5.5 with 100-round editing, and a $2.80 synthetic influencer drew almost 1,000 bookmarks.
Engagement figures were read from X, YouTube, Hacker News and the Unreal Engine forums between about 2:45 and on Nov 9 and will keep changing. X search was not reachable this run, so X posts come from the public timelines of the accounts we follow for this edition. Anything outside that roster is missed. Reddit refused our requests this run.
⭐ Pick of the day
💻 Code and UI
- Claude Fable 5.2 is out, and the first design tests are already 3D: a 30-minute build at 45,536 views, Blender plus Reaper sound design, and a code-only jet (Anthropic, @LexnLin, @AmirMushich), news. Anthropic's launch page (dated Nov 9) says Fable 5.2 and Mythos 5.2 are "the same underlying model," with Fable the version for all users, and that it generates "up to 30% faster" than Fable 5.1 at $8/$40 per million input/output tokens, down from $10/$50. The design-relevant claims are Anthropic's own and we did not test them. In one test, a one-paragraph prompt led the model to design three working devices (a voice-assistant hub, a USB speakerphone and a handheld virtual pet), routing their circuit boards and modelling 3D-printed enclosures in CAD. On BenchCAD 2.1, which the page describes as a hardware-engineering benchmark, Anthropic reports 83.5 at max effort against 65.3 for Opus 5.5. A quote from Unity on the page says it builds "higher-quality 3D scenes from scratch" and rebuilds "an experience from a visual reference." The page also says that for most coding work users should start with Opus 5.5, its "strongest daily driver." The signal: the launch reached 797 points and 470 comments on Hacker News, and Artificial Analysis's post ranking it #1 on its Intelligence Index, at 65, has 1,722 likes and 107,328 views. That post also gives an average cost per task about 2.6 times Opus 5.5's at max effort. The makers: @LexnLin's clip (), "this took ONLY about 30minutes with Fable 5.2 and a few opus 5.5 subagents," has 973 likes, 295 bookmarks and 45,536 views. It names no tools beyond the models, and we could not see how it was made. @AmirMushich tested it with Blender and Reaper (, 60 likes, 14,697 views), recording "most sounds on my iPhone" and handing them to Claude. They praise the sound design but say they are "not a fan of the melody." Promotional: Higgsfield's jet () claims "931 parts. 24,260 rivets. All modeled in code in Blender" via its own MCP. Higgsfield is also running a free-trial offer around the launch, so treat its demos as marketing. Why people are losing it: within hours of release, makers are using a new frontier model as a Blender operator, not just a chatbot. How to try it: if you have a paid plan or API access, start with a small brief (one object, a fixed part count, one camera move) in a scripted Blender session. Per Anthropic's own advice, keep Opus 5.5 for routine work.
The ranked picks
🧊 3D and games
- Unreal Engine 5.9 MegaJam Preview: Claude Code, Codex and Cursor inside the editor, agents that install Fab assets, and Unreal Canvas for AI images, video and meshes (Epic Games), news. Epic's announcement post () has 1,224 likes, 444 bookmarks and 92,114 views, and the forum post shows 6,416 views. Per Epic, this is "the first Unreal Engine release built primarily with agentic workflows." It was put together in about a week for the Epic AI MegaJam (Nov 12 to 15), had "a shorter QA pass than usual," and parts of the agent-written web renderer are prototypes. All the new features are marked Experimental. Unreal MCP now starts with the editor and can search Fab and install assets. Unreal Assistant runs your own coding agent in an editor panel. Unreal Canvas, a web app and editor tool, generates concept art, images, 3D meshes and video from prompts, references "or your own scenes while preserving composition and camera perspectives," using models you connect with your own API keys. Projects can also run in a browser on WebAssembly and WebGPU, though that needs a source build. @Stefan_3D_AI's summary (566 likes, 23,579 views) says Epic's Canvas demo clip uses Nano Banana Pro, Tripo 3.1, Seedance 2.0 and a local Flux restyle. That list is their reading of the clip. Why it's here: the biggest game engine just put agents and generative tools inside the editor, days before a jam built around them. How to try it: per Epic, install it on a copy of a project, never your live one.
- AI video on the left, a Blender rig on the right: "Video vs Opus 5.5 after 10 iterations" (@dhruvalgolakiya), showcase. Posted at : 186 likes, 92 bookmarks and 22,404 views. The post is one line plus a clip, and its preview frame labels a costumed gunslinger "AI video" on one side and "Blender rig" on the other, so the claim is that Opus 5.5 rebuilt a generated character as a rigged Blender model in ten rounds. We did not see the method or files. This maker was in Nov 7's editions with a different project. Why it's here: it is the clearest side-by-side we saw today of an agent turning a video character into an editable 3D asset. How to try it: give the agent a still from your clip and ask for a blockout first, then a rig, comparing against the frame after each pass.
- KetchUp, a free, open-source SketchUp alternative built with Rust and WebGPU, draws 726 points on Hacker News, tool. Posted on Nov 8 at , it has 726 points and 222 comments. The page describes free 3D modelling for woodworking and furniture on Windows, macOS, Linux or in the browser, and now titles itself "BenchUp." The AI angle comes from the comment thread, not the page: commenters treat it as one of a wave of agent-built clones. One counts about 161,000 lines of code, and another says it "can't extrude a cube without glitching." We did not confirm how it was built. Why it's here: it is the most-discussed design tool of the day, and the thread is a frank debate about agent-built creative software. How to try it: open the browser version and model one simple piece of furniture before you trust it with a real plan.
🖼️ Image
- ⏪ update: Seedream 5.5 officially launches, with 100-round edits that leave untouched areas "intact at the pixel level" and native transparent layers (ByteDance Seed), news. Today's noon edition carried one maker's first test. This is the official launch: ByteDance's page claims multi-turn editing "keeping unedited regions intact at the pixel level" and "native support for transparent and semi-transparent backgrounds," including outputting each character or sticker as its own transparent image. TestingCatalog's rollout post (, 247 likes, 12,285 views) says it is arriving on Dreamina, CapCut and BytePlus. Other makers' posts repeat "100 rounds" of edits, which is ByteDance's claim, not a test we ran. Why it's here: it is the day's main image-model launch, and transparent output is something designers ask for constantly. How to try it: change one thing per round on a single product shot, and check that the background stays put.
🎬 Video and animation
- A "yapping influencer" that "costs $2.80" and "doesn't even exist," made with Wan 3.0 Prime and a character sheet (@simonecanciello), workflow. Posted at : 616 likes, 996 bookmarks and 68,369 views, with bookmarks well ahead of likes. The cost and "i made her in 5 minutes" are the maker's claims, and the workflow is offered on request, not posted. The maker has posted ad-platform promotions before (in Nov 8's noon edition). Why it's here: it is the most-saved video post we read today, and it is exactly the kind of synthetic person that platforms ask creators to label. The maker adds, "i'm gonna get scammed when i'm old." How to use it: if you make one, disclose it as AI-generated on every post.
- "Augmented Future," a supermarket AR short in Midjourney Niji 6 and Seedance 2.5 (@jboogx_creative), showcase. Posted at : 220 likes, 90 bookmarks and 5,769 views. The setup is a shopper reaching into the freezer when "the second scene happens," and the post ends with "What do you do?" The maker names the two tools and nothing else. They were in Nov 8's noon edition with a different piece. Why it's here: it is a well-liked short built on a story beat, not a template prompt. How to try it: design the key frames in Niji, then animate only the moment the scene changes.
💻 Code and UI
- Skill lists for design agents: 20 skills for video, design, 3D, motion and gamedev (@Y0urDumble), and 8 free product-design skills (@rpnickson), workflow. Y0urDumble's list () has 297 likes, 607 bookmarks and 25,154 views. It groups repos such as HeyGen's hyperframes ("write HTML, get a video"), OpenMontage and the official Remotion skills, and it gives star counts that are the poster's figures ("770K+ github stars between them"). The opener "just cooked a 20-person motion studio" is hype. rpnickson's thread (, 89 likes, 126 bookmarks) argues that out-of-the-box agent designs "start looking the same" and lists free skills such as Stageshow, Impeccable and Dimensional Drawing. Both makers appeared in today's earlier editions with other posts. Why it's here: two lists aimed at the same complaint, the "Claude look," and together they drew more than 700 bookmarks. How to try it: install one skill, run the same brief with and without it, and keep it only if the output changes.
Method: X search could not be reached (x.com refused the browser), so X coverage comes from the public timelines of an 820-account roster read through a public read-only mirror: 2,943 posts from the last 30 hours, filtered by keyword. Every linked post was re-read by ID for its text, time and metrics, and media previews were viewed where the text was thin. The sweep also covered Hacker News (stories from the last 30 hours with comments read for the two linked threads), YouTube searches limited to today's uploads (with watch pages read for dates and descriptions), the Anthropic, ByteDance Seed and Unreal Engine forum pages, and web search. Picks run best first within each medium, by engagement and velocity, how many people are copying them and how new they are. The ⏪ item (Seedream 5.5) is a deliberate update on a noon item, with official launch news. Left out as repeats from today's morning and noon editions: the image-to-sculpt skill, Recto, the MiniMax H3 multi-character swap, NVIDIA's Open-H3-Regenerate-2K refiner and its ComfyUI nodes, the Opus 5.5 news-video edit and @techartist_'s 3D anatomy book. Also left out: Higgsfield's free-trial launch post (promotional, 1,012 likes), "Fable 5.2 built an operating system" repost threads, model-leak and pricing posts with no design work, celebrity-likeness Nano Banana prompts, the "This AI video is a masterpiece" Seedance template posts, money-making Claude Code threads, and a 7-figure startup sale. Not read this run: Reddit (requests refused), GitHub trending, Hugging Face and Product Hunt.
Follow every creator we feature on X: Daily Design Creators on X
Say the word and I'll steer the next edition there.