Rodin 3D Gen-2.5 - Text to 3D

Text-to-3D is the glamour task of this whole category - no reference, no sketch, just words and a hope - and Rodin Gen-2.5's text node is one of the most capable freeform versions you'll get inside ComfyUI. Same generation line as the image variant, same mode switch, but the input is a prompt instead of pictures, so the whole game becomes how well you can describe a three-dimensional thing in a single text box.

The familiar fine print: paid cloud API node, Comfy Org account + credits, no local compute. And the familiar text-to-3D honesty: quality runs behind image-to-3D, and Rodin's own service will say the same thing. This node is for when you have no reference and need a first draft to react to - not for hero assets.

How it works

Type a prompt, pick a mode, and the node uploads the request to Rodin's Gen-2.5 API via Comfy's cloud. The same dynamic mode combo applies as the image node: Fast (1K–20K faces, quick drafts), Regular (balanced, with tier and polygon controls), and Extreme-High (20K–2M faces with optional micro-detail). Comfy polls until the model is done and downloads it into output/Rodin3D_Gen25_<uuid>/ in your chosen format.

The inputs that matter

• prompt - the whole ballgame. Specific and material-rich wins: "an antique brass compass with a cracked glass face and a tarnished silver chain" gives the model geometry to infer; "a compass" gives you a generic blob.
• mode - Fast for silhouettes, Regular as the default, Extreme-High only for finals. The price badge tracks this.
• material - PBR, Shaded, All, or None (default Shaded). PBR gets you game-engine maps; Shaded is easier on the eyes in a viewer.
• geometry_file_format - glb (default), fbx, obj, or stl.
• Optional: texture_mode, seed, TAPose (for human-like results), hd_texture, texture_delight, addon_highpack, plus bbox_width/height/length and height_cm to constrain size.

Output: a single model_file (FILE_3D) in your chosen format. One-shot, no task ID to chain.

Common issues

The disappointment curve is real: text-to-3D will surprise you sometimes and embarrass itself other times, and no seed will save you (non-deterministic, per the tooltip). Don't pay for Extreme-High on a draft - you're buying resolution on geometry that may be wrong. And remember the mode's hidden costs: addon_highpack (4K textures, ~16x faces) is the pricey add-on, and bbox/height constraints are free but only apply when you actually set all three. If the result is close but rough, that's what an image-based refine pass on the same service is for.