Rodin - Gen 2.5 Regular Image-to-3D

The Rodin - Gen 2.5 Regular Image-to-3D node is the balanced option in the pack's 2.5 lineup, and it's the one I'd tell a beginner to reach for first. Fast tiers are for drafts, ExtremeHigh is for the final hero asset, and this sits in between - you get a real polygon budget (up to a million faces if you want them), a quality-tier dropdown, and a turnaround that won't make you abandon the whole idea. The name says "Regular." It's not a diss; it's positioning.

What it is

Image-to-3D on Rodin's Gen 2.5 model: you feed one or more images, it reconstructs a textured mesh with PBR or shaded materials. Like every node in this pack it's a cloud API wrapper - your image tensors get uploaded to https://api.hyper3d.com/api/v2/rodin, the API generates server-side, and the node polls until the mesh is ready, then downloads it to ComfyUI/output/<timestamp>/. Your GPU is a spectator. The README frames the whole extension around the Rodin API, and this is its bread-and-butter mode.

The inputs that matter

Most of the required set is the pack-wide standard (api_key, images, seed_, Material_Type, TAPose). The two that actually change your result:

• tier - Gen-2.5-Low, Gen-2.5-Medium, or Gen-2.5-High (default High). This is the price/quality lever: Low is cheap and quick, High is the good stuff. There's no shame in starting Low to validate an idea.
• Polygon_count - the big one. Options run from 4K-Quad up to 1M-Triangle, plus Default, which delegates to whatever your chosen tier recommends (Medium/High default to a 500K raw mesh). "Quad" gives you subdivision-friendly quad topology; "Triangle" gives raw triangulated meshes. If you're going into a game engine, pick a quad preset and never look back.

geometry_file_format is glb or usdz here (the fancier formats are reserved for ExtremeHigh), and texture_mode ranges from extreme-low to high with a Default that defers to the tier. Optional bbox and height_cm let you pin physical size - wire in the pack's BBox Controlnet node if you want sliders for it.

What comes out

One output, model_path - a string path to the downloaded mesh. Wire it into ComfyUI's core Preview3D node (model_file input) to inspect the result before you do anything with it. That's the pack's own example-workflow pattern.

How it compares to its siblings

• Fast Image caps mesh faces at 20K and defaults to the Low tier - that's the iteration node.
• Regular (this one) lets you scale all the way to 1M faces and pick topology. Draft or deliverable, your call.
• ExtremeHigh adds mesh modes, face counts to 2M, five output formats, and a Creative toggle. Better, slower, pricier.

If you're unsure which node to put in a workflow you're still building, this is the safe default.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/DeemosTech/ComfyUI-Rodin.git
pip install -r requirements.txt

or "ComfyUI-Rodin" via ComfyUI Manager, then restart. Dependencies are aiohttp (the HTTP layer) plus asyncio, which is stdlib - the pack's requirements line for it is a harmless no-op. No models to download, nothing heavy.

Troubleshooting

Same family as the rest of the pack: MISS_API_KEY means the key node isn't connected or is empty; a wrong or creditless key surfaces as an API auth error at generation time; the 30-minute polling cap can expire on big jobs, in which case the console prints a timeout and you should check the Rodin dashboard - the job may have finished and just not downloaded. If you're iterating and jobs feel slow, remember the tier is the knob. Draft on Low, ship on High.