scanfi @GuideboardLabs

I built an AI agent that can turn a single reference photo into a 3D model in Blender. No manual sculpting, no GUI tweaking. Just code, render, feedback, repeat. Using this Computer Use update. The agent runs on Hermes from @NousResearch. I drop in a reference image, the agent analyzes the proportions through a vision model, writes a Blender Python script, renders it headless at 1920x1920 in about two seconds, and I give feedback. It iterates until the model matches. The technical path was the interesting part. The agent learned to use bmesh for precise topology instead of relying on modifiers for everything. The tapered stackable base needed explicit vertex ring placement. The D-shape handle came from a four-point Bezier curve converted to mesh and merged into the body with a boolean union. That handle was the hardest piece. A Bezier tube's flat end cap against a curved cylinder always leaves a visible gap. Converting to mesh and using boolean union was the fix. The modifier stack order mattered too. Solidify, then Bevel, then Subdivision Surface. Bevel after subdiv creates artifacts. Learned that one the hard way. The infrastructure is all local. The agent runs on an AMD 5700 XT box with a cloud vision model for image comparison. The vault is searched before every response for relevant project knowledge through the OverCR bridge, which gives persistent memory across sessions. That continuity is what made the iteration loop work. Each session picked up where the last one left off. The broader point is that an LLM with vision, code generation, and the ability to check its own output can become a competent 3D modeler. Not through a plugin or a fancy interface. Through writing code, rendering, getting feedback, and iterating. That pattern applies to any domain where you can describe what you want and verify the result. He’s at your desk ready to create.

Cua @trycua

2 weeks ago

1/ Today we're bringing Cua Driver to Linux: background computer-use for any agent. Hermes, Claude Code, Codex, or your own loop can drive real Linux desktop apps (X and Wayland in preview) through CLI or MCP while your desktop stays usable

46 98 781 129K 616

1 1 4 233 1

View Details

scanfi @GuideboardLabs

a day ago

The future was yesterday

Fraser Price @fraserpricee

2 days ago

omg please regulate me daddy @GOVUK its too dangerous please we need to do something its inside my computerrrr it might learn to use microsoft excele or run as an MP we need to do id verification to download a gpus we need to save the kids thank u for ur attention to this matter

9 2 123 24K 31

0 0 0 50 0

View Details

scanfi @GuideboardLabs

a day ago

@jpfraneto @NousResearch @stripe @nvidia Temp tunnel with CF …amination-von-sanyo.trycloudflare.com My domain on IONOS has to wait 24 hrs to do a DNs reset so can’t use it.

0 0 0 22 0

View Details

scanfi @GuideboardLabs

2 days ago

I built an autonomous merch agent on @NousResearch's Hermes Agent for the @stripe & @nvidia hackathon Welcome to Tiki Hermes, the digital boardwalk hub for gift shop merch It scans X in real-time, designs T-shirt concepts from what's trending, and sells them through a live Stripe storefront. The agent does the creative work. Rendering is fully automatable too, I chose manual upload, but have Gemini backend available. Everything runs on its own, with you in the loop if wanted. 🧵👇

Nous Research @NousResearch

2 weeks ago

The Hermes Agent Accelerated Business Hackathon presented by @NVIDIAAI × @stripe × @NousResearch starts now, for builders making agents that can earn, spend, and run real operations at any scale. Our NVIDIA integrations let your team run agents safely through NemoClaw, quickly

182 241 2K 1.8M 1K

8 0 5 424 0

View Details

scanfi @GuideboardLabs

a day ago

@jpfraneto @NousResearch @stripe @nvidia Lmao yeah my agent started that for me but I had some other stuff I was looking into. Will have him wrap that up now

0 0 0 14 0

View Details

scanfi @GuideboardLabs

a day ago

@macbethAI @dillonrolnick @NousResearch @UnrealEngine Did you need to set up assets or did you just use the giant library of 3D scans they have?

0 0 0 13 0

View Details

scanfi @GuideboardLabs

a day ago

@charles_maddock GLM 5.2

0 0 1 150 0

View Details

scanfi @GuideboardLabs

a day ago

@tamanekokoro DeepSeek V4 Flash has been my gateway GOAT for awhile now. Spawns most jobs to Qwen3.5-9b tasks.

0 0 1 232 0

View Details

scanfi @GuideboardLabs

a day ago

@Teknium Great work guys

0 0 3 184 0

View Details

scanfi @GuideboardLabs

a day ago

Who is missing?

0 0 1 38 0

View Details

scanfi @GuideboardLabs

a day ago

Storefront live over tunnel @ tikihermes.loca.lt Might port later @steipete we’ve got one fresh off the presses for you

scanfi @GuideboardLabs

2 days ago

8 0 5 424 0

0 0 1 62 0

View Details

scanfi @GuideboardLabs

a day ago

@jpfraneto @NousResearch @stripe @nvidia tikihermes.loca.lt Might throw it on my dotcom if enough interest is there

1 0 1 33 0

View Details

scanfi @GuideboardLabs

2 days ago

@fagamericano @Felirami @steipete @openclaw It’s mostly karma for his constant sneak dissing at competitors. And it’s honestly embarrassingly bad. Why would you trust an agent that when used by its creator comes up with that…

0 0 2 117 0

View Details

scanfi @GuideboardLabs

2 days ago

Built using @Teknium Hermes Agent, for all code and planning (even made the demo video) Full demo video in the replies. Storefront is live. Stripe checkout processes real test payments. github.com/GuideboardLabs (repo coming soon)

0 0 0 39 0

View Details

scanfi @GuideboardLabs

2 days ago

Most AI merch tools pick a topic from a list and slap text on a shirt. Boardwalk's agent actually reads what's trending on X, reasons about what makes it funny, and designs concepts with full art direction. The concepts are timely, specific, and have visual jokes. The pluggable architecture is the point. Swap scanners, swap renderers, swap products.

0 0 0 22 0

View Details

scanfi @GuideboardLabs

2 days ago

Two servers, zero exposed attack surface: • Storefront: Flask app, public URL, Stripe checkout. No admin endpoints exposed. • Admin: Separate Flask app, private port, Tailscale-only. Dashboard, upload, scan triggers, Stripe key management, transaction viewer. The agent drives the pipeline. The servers are just dumb endpoints.