Chaining four edits in Nano Banana 2.1: most of the room survives
AKAMA199 · reddit · 2026-10-27
A Reddit user chained four successive edits on the same Victorian room with Nano Banana 2.1 Edit (via atlas cloud), feeding each result into the next round:
- Floral throw → faded blue knitted blanket; portrait eyes → black buttons; framed photo → handwritten WELCOME HOME note; two Santa figures → handmade button-eyed dolls (Coraline-style).
Surprisingly, the fireplace, curtains, table, wall color, lighting, and rug mostly stayed put after four chained edits, and earlier edits carried through later rounds.
Lessons learned:
- Button eyes needed a more explicit instruction than expected;
- Replacing the Santas was hardest — the first attempt left them alone until the prompt explicitly demanded removal of the originals;
- The dolls came out more stylized than the rest of the image.
The author posted the original, all four stages, and a side-by-side comparison.
More from Multimodal
- KlingAI study: visual memory boosts consistency in long-video generation but can lock in outdated appearances — Yuzhuo Chen · 2026-10-27
- CausalAlign aligns video generation to 3D dynamics, boosting geometric consistency over 20% — Kai Zhu · 2026-10-27
- Media.io AI video hides a "twist" in a photorealistic cliffside wave scene — Stephane_AI · 2026-10-27
- Dreamina's $0.29 tier produces surprisingly detailed, hard-to-fake human video — SignificantMode6371 · 2026-10-27
- Z-Image Turbo turns one: the 6B model that beat Qwen Image at realism — ReferenceConscious71 · 2026-10-27
- Realistic "dumping stuff into a volcano from a plane" video made on Dreamina's 29-cent plan — Ok-Possible-5566 · 2026-10-27