meshy / v6 / textTo3D

Text-to-3D is the awkward cousin in this whole space: in the open ecosystem it's almost never a native model, just text-to-image followed by image-to-3D wired into one graph. Meshy is one of the few places that runs a genuine text-to-3D path, and this node puts it in your ComfyUI graph, billed in Buzz. If you want a character from a description and you don't want to hand-draw the reference image first, this is the one you reach for.

What it is

meshy / v6 / textTo3D from civitai/civitai-comfy-nodes, Civitai's official pack for its Orchestration API. The required input is a plain prompt - a multiline STRING, your text description - and a hosted Meshy v6 job turns it into a mesh. No reference image needed, which is the whole point.

The inputs

• prompt - required. This is your entire creative input, so it deserves your attention. Describe the object, its style, materials, and any character posture; Meshy is noticeably better when the prompt is specific.
• mode - preview or full, default full. Preview mode is the cheap, fast scout - run it to check the concept, then spend on full once you're happy.
• enable_prompt_expansion - off by default. Turn it on to let Meshy rewrite and enrich your prompt before generation. Good for vague prompts, mildly annoying if you want exactly what you typed.
• target_polycount (30,000), topology (quad/triangle), symmetry_mode (off/auto/on) - the mesh-shape controls, same as the imageTo3D variant.
• should_remesh, should_texture, texture_prompt, enable_pbr - pipeline toggles. texture_prompt is your chance to steer materials separately from the geometry description.
• enable_rigging, enable_animation - off by default; the reason to be on Meshy rather than the comfy nodes. Rig a character, even animate it, and get usable rigged_model / animated_model outputs.

The outputs

model (GLB URL), fbx_model, thumbnail (IMAGE), rigged_model, rigged_fbx_model, animated_model, animated_fbx_model, basic_animations (populated when rigging/animation are enabled), plus workflow_id and raw_json.

Installing it

ComfyUI Manager → Civitai Comfy Nodes → Install → restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Gotchas

• Describe, don't dash. A one-word prompt on a text-to-3D model is a lottery ticket. Structure it: subject, pose, style, materials.
• Use preview mode first. mode: preview exists exactly so you don't pay full price for a concept you'll reject. Run preview, then full with the seed and polycount you like.
• Billed in Buzz, per run. This is a commercial Meshy job routed through Civitai; raw_json shows the cost, and enable_animation is the most expensive single toggle.
• Credentials required - Civitai Auth node, CIVITAI_API_TOKEN, or sidebar OAuth.
• Realistic expectations. Hosted Meshy output beats the open stuff on characters, but it's still AI-generated topology. Retopo for anything that has to deform in production.