- Authoring time
- 4m 25s
- Staged build time
- 40.9s
- Total tokens
- 572,903
- Est. API cost
- $0.2628
Editor’s pick · GPT-6 Astra
Classic Mammoth Tank
One prompt. Different AI models.
@Zachas
A heavy twin-cannon Mammoth Tank inspired by classic Command & Conquer. Compare tracks, armor, missile pods and detail across independent Blender builds.
Drag to rotate · Scroll to zoom · Pinch on mobile
Compare the model runs
Same prompt. Different approaches. Inspect the results and the work behind them.
- Authoring time
- 4m 13s
- Staged build time
- 27.8s
- Total tokens
- 674,641
- Est. API cost
- $0.0153
- Authoring time
- 5m 43s
- Staged build time
- 51.3s
- Total tokens
- 252,053
- Est. API cost
- $0.9078
Original prompt
# Identical model brief Create a detailed, immediately recognizable classic Mammoth Tank inspired by the original Command & Conquer in Blender 5.1.2. Exterior display model only: very heavy low armored hull, two broad caterpillar track assemblies with visible tread links and wheels, a substantial rotating-looking turret with twin parallel cannon barrels, twin missile pods, hatches, vents, headlights and restrained faction-color accents. Capture the chunky retro real-time-strategy silhouette and weathered military metal, in a polished stylized low-poly diorama on a small terrain display base. Cannons and tank front face negative Y. Keep the whole tank visible and centered. This is decorative fictional game art; no functional weapon design or technical internals. No downloaded assets. Decide the geometry, proportions, details and artistic interpretation independently. Deliver scene.py in your assigned model directory. It must expose def build():, a Python generator using bpy that creates the scene from an empty scene and yields a short English stage description after each meaningful increment. Aim for 18–35 increments so an actual Blender window recording shows construction. No sleep, no rendering, no saving, no quitting Blender; the shared harness handles those. Module import must not create geometry. Use only bpy/mathutils/Python standard library. All scene geometry should fit approximately inside x/y [-9,9], z [0,11]. The subject faces negative Y. Assign material diffuse_color as well as node colors for visible material colors in solid viewport. Include a camera looking from (14,-19,13) toward (0,0,3) and suitable lights. Keep total objects below 2500 and avoid expensive modifiers. Use Blender 5.1 API. Generator execution happens in Blender UI via timer, so do not rely on area context for geometry. Do not modify UI or viewport; harness does that. Own only your assigned model directory. Other workers may be present; do not revert or inspect their work or any other model's result. Do not use other AI models. Do not launch Blender, take over windows, install anything, or run an independent creative build. You may check Python syntax. Parent harness runs your actual Blender build and will report execution errors for repair. Write a concise design.txt identifying your intended result. End after these files are ready. Same reasoning setting for all models: high. Each is a separate fresh agent. Timing separates model-authoring turnaround, actual Blender generator execution, visible staged build wall time, and common setup/QA. A single run is an exploratory comparison, not a statistically reliable benchmark.