When AI Can’t Count: 8 Solar Terraforming Vehicles + 15-Scene Cube Story (Limit Case Study)

[silent furnace]

I ran into an interesting limitation while creating two complex image concepts:
A single wide shot that must contain exactly eight photovoltaic “terraforming” vehicles (excavator, PV van, PV trailer, concrete mixer, lift platform, tractor, solar fire truck, small trailer). The model often generated 7 instead of 8, merged objects, or dropped one during composition.
A storyboard made from five 3-face cubes — meaning 15 distinct scenes (3 per cube). Even with numbered prompts, the model struggled to keep every small panel unique and consistent, especially at miniature scale.
This isn’t just prompt skill — it shows a real constraint around object counting, layout planning, and multi-panel composition. My workaround was an iterative workflow (generate parts separately, then assemble), but it would be amazing to have future tools that support layout-first planning, hard constraints (8/8 present), and panel-aware generation.
These images matter to me because they visualize a clean-energy “terraforming” team and a found-object storytelling method — both are part of a bigger vision for tech that helps the planet.

ImageGen PromptEngineering FailureCase LayoutControl ObjectCounting Storyboard ComicLayout AssemblageArt SolarVehicles TerraformingPlanet ClimateTech CleanEnergy CreativeTools AIArt Iteration

Click here to read the full suggestion & discussion

[silent furnace]

I have a problem setting the angle of the upper part of the solar umbrella so it aligns with the light, even after hand-drawing the design and manually marking all the components.

[silent furnace]

Task definition:
6 cubes, each showing 3 visible faces, numbered from 1 to 6. Each cube must contain exactly three different letters, one per face, following the alphabet in order:
Cube 1: A, B, C
Cube 2: D, E, F
Cube 3: G, H, I
Cube 4: J, K, L
Cube 5: M, N, O
Cube 6: P, Q, R
What went wrong: even after a detailed explanation, the generated image still breaks this exact mapping. Letters are duplicated and others are missing (H missing in cube 3, L missing in cube 4, O missing in cube 5). This shows the same logical error repeating despite clarification.
What I would need to fix in myself: I need to slow down and reason explicitly per cube instead of assuming the sequence holds. I should mentally validate each cube one by one (Cube 1 A/B/C → Cube 2 D/E/F → Cube 3 G/H/I, etc.) and then verify all 18 faces before presenting the result. I must not trust the generator’s text handling and should apply a strict internal checklist and longer reasoning pass before claiming correctness.
Bottom line: the issue is not unclear instructions — it’s my lack of strict verification and focus. To avoid repeating the same mistake, I need to think longer, validate systematically, and only respond once the full cube-by-cube logic is confirmed.