Replicated exact original generation setup for each reference image
Each group uses the same model, refs, and prompt structure as the batch that created its reference
Groups A-D: NBP with 3 refs (O2 style + Ollie + Dot)
Groups E-G: NB2 with 2 refs (Ollie + Dot only, no style ref)
Group H: NBP with different style ref (d3_in_motion, not O2)
Only the scene/character dynamics changed to hide-and-seek
Zero mixed media language, zero added material blocks
Batch 2 (Failed — Redoing)
2026-05-13
Founder Feedback
Images came out in mixed media collage style — completely wrong
Reference images are 3D photographic style, NOT mixed media
Root cause: prompt contained material block ("torn paper coral," "gouache," "watercolor paper base") that was NOT in the original batch22/23 prompts
Need to use the EXACT SAME prompting and API calls that created the reference scenes
Action
Batch 2 regenerated using exact batch23 ROLE_BLOCK + CHARACTER_BLOCK with zero mixed media language. Only the scene paragraph changes to hide-and-seek.
Batch 1
2026-05-12
Founder Feedback
Results aren't really working
All sub-batches used essentially the same reference images — didn't leverage the variety of scene images provided
Need to use each provided scene image AS the style ref, not as a scene composition ref
Need varied prompts adapted to each reference image's mood/style
Generation info showed "(Same prompt as A #1)" — need full unique details per image
Favorites and Export Favorites not showing properly
Action
Batch 2 generated with 10 different style refs from founder's reference images, each with unique tailored prompt