Use the 3D AI Studio MCP to run the "Stylized Robot & Mech" template (https://www.3daistudio.com/templates/robot-3d-model) for me.

My idea: A heavy battle mech with shoulder cannons
Build: As described
Condition: New

Follow these steps in order. The prompts and settings are the template's own, so use them word for word and only put my idea where it says {MY IDEA}.

1. Design the asset
Call generate_image_nano_2_1 with aspect_ratio "1:1", resolution "1K", output_format "png" and this prompt:
"""
A finished 3D game asset render of {MY IDEA}. Style: stylized sci-fi 3D game asset, clean hard-surface panels with bevelled edges, a bold two-tone paint scheme with small accent colours and readable shapes, like a modern stylized space game. One single robot or mech, full body from head to feet, standing upright on its own feet in a neutral, readable stance with arms slightly away from the body, facing three-quarter front, weapons and tools attached to the arms or back, the entire robot visible with generous empty margin. Seamless pure white (#FFFFFF) background, soft even studio lighting, no cast shadows, no ground plane, no smoke, no muzzle flashes, no text. Surface and light: soft, even, diffuse studio light. Metal, gems, glass-like and wet or glossy materials show what they are through colour and a soft satin sheen, never through mirror reflections, bright specular hot spots or reflections of the surroundings. Glowing parts (eyes, runes, lava, neon, gems) are painted as bright, flat emissive colour on the surface itself, with no light spilling onto nearby surfaces, no halo, bloom or glow around the object, and no light rays, sparkles or particles in the air.
"""
If I want to start from a sketch or reference image instead, call upload_image so I can hand it over, then call edit_image_gpt with the upload in images, quality "low", output_format "png" and this prompt:
"""
This is a rough sketch. Recreate the object it shows as a finished 3D game asset render, keeping the sketch's design, proportions and key details, and ignore any handwritten notes, arrows and paper. Style: stylized sci-fi 3D game asset, clean hard-surface panels with bevelled edges, a bold two-tone paint scheme with small accent colours and readable shapes, like a modern stylized space game. One single robot or mech, full body from head to feet, standing upright on its own feet in a neutral, readable stance with arms slightly away from the body, facing three-quarter front, weapons and tools attached to the arms or back, the entire robot visible with generous empty margin. Seamless pure white (#FFFFFF) background, soft even studio lighting, no cast shadows, no ground plane, no smoke, no muzzle flashes, no text. Surface and light: soft, even, diffuse studio light. Metal, gems, glass-like and wet or glossy materials show what they are through colour and a soft satin sheen, never through mirror reflections, bright specular hot spots or reflections of the surroundings. Glowing parts (eyes, runes, lava, neon, gems) are painted as bright, flat emissive colour on the surface itself, with no light spilling onto nearby surfaces, no halo, bloom or glow around the object, and no light rays, sparkles or particles in the air.
"""

2. Build the 3D model
Call generate_prism_3_1 with the task_id of the image from step 1 in images, texture_model "3.5", texture_quality "fast", geometry_quality "standard", pbr false, face_limit 10000, lighting_removal false.
For a lighter, game-optimized mesh with clean low-poly topology, call generate_p2 instead with texture_model "3.5", texture_quality "standard", pbr false, lighting_removal false and no face_limit.

3. Export
Export it for Unity unless I name another target: call convert_model on the 3D model with output_format "fbx_zip" (FBX with its textures) for Unity, Unreal / UEFN or Roblox Studio. Godot or Blender take the generated GLB as it is, no conversion needed.
For Roblox, the mesh has to stay under 8,000 faces: pass face_limit 8000 to the 3D step.

After each generation call, keep polling get_generation with its task_id until it completes, then show me the result with show_generation before you start the next step.