I had shown how to create these pictures with 2 different prompts, but someone asked me to merge the 2, so here is how you can do it in less than 2 minutes with NanoBanana. 🤗

📝 TL;DR Luma just launched Ray3 Modify, an AI video model that lets you radically rework real footage while keeping the original performance intact. Think wardrobe swaps, new locations, and full scene redesigns, all built on what the actor actually did on set. 🧠 Overview Ray3 Modify is a new set of tools inside Luma’s Dream Machine that combines AI video generation with real cameras, real actors, and real performances. Instead of starting from text prompts, you film once, then use AI to transform wardrobe, environment, lighting, and even character identity while preserving motion, timing, and emotion. It is aimed squarely at filmmakers, brands, and creators who want high end control without high end reshoot costs. 📜 The Announcement Luma announced Ray3 Modify as a new hybrid AI workflow for acting and performance, now available in Dream Machine. The update brings three headline capabilities, Modify Video, Modify with Keyframes, and Character Reference, all designed to keep human performance at the center while AI handles the heavy lifting on visuals. The goal is to fix a core flaw of early AI video, it was expressive, but could not reliably follow or preserve what actors actually did on camera. ⚙️ How It Works • Modify Video - You upload a clip, then use Ray3 Modify to change wardrobe, environments, lighting, and product placement while keeping the physical logic, narrative flow, and performance authenticity of the original shot. • Modify with Keyframes - You can set Start and End Frames to guide longer moves and transitions, telling the model how the scene should begin and end so it respects camera motion and spatial continuity in between. • Character Reference - You add a character reference image, then Ray3 Modify projects that identity onto the actor, locking likeness, costume, and identity continuity across the whole clip. • Performance preservation - The model is conditioned on the original footage, so it follows the actor’s real motion, timing, eye line, and emotional delivery while transforming the look of the scene around them.

After logging 300+hours of intensive AI use in the last six weeks, after the AI bootcamp, 40 days ago from Nov 6th, 2025, running real workflows—not demos—I made a decisive platform shift to GEMINI PRO. Below is the exact rationale, from my personal and practical use documented for transparency and repeatability at the request of some of my followers. ChatGPT — Limitations Observed in High-Intensity Use (300+ Hours of Practical Work) ✘ Inconsistent reliability during peak usage hours; frequent bottlenecks ✘ Constantly have to reopen new chats to continue an existing workflow ✘ Consistently forces the user to refresh the chat when it crashes ✘ Uses DALL·E for image generation (Inferior MCP connector) ✘ Very poor quality in image AI generation ✘ Uses report PDF generator, which is inaccurate and does not parse text properly ✘ Does not produce native video outputs in comparison to Gemini Pro VEO 3 or upcoming VEO 4 power ✘ Does not integrate like Gemini Pro with Google Videos, Products, and Workspace ✘ Does not check Gmail inside the chat interface ✘ Uses folders, but does not use Gems like Gemini Pro, which are sub-agents with rules, (Time Saver) Gemini Pro — Why I Transitioned (Now Operating as “Nautilus”) ✔ Exported ENTIRE history of conversations from ChatGPT to Gemini Pro very easily. ✔ Stronger long-context persistence across extended conversations ✔ Stable performance during sustained, high-intensity usage ✔ Supports structured agent orchestration with Gems ✔ Enables purpose-built autonomous agents (“Tritons”) ✔ Reduces prompt redundancy through retained role-specific logic ✔ Scales horizontally without performance collapse ✔ Better aligned with real business, automation, and execution workflows ✔ Superior image creation with Nana Banana Pro ✔ Superior creation of videos with VEO3 & VEO4 ✔ Checks Gmail inside Gemini Pro chat interface (Genesis Pro user face) ✔ Pulls YouTube videos directly inside the UI chatbox, which GPT does not do ✔ Grabs direct research from Google, which ChatGPT cannot do

Automation is a force multiplier. That’s the point… and also the risk. It’s genuinely “survival” once volume and response-time expectations hit. But automating too early (or automating a messy workflow) doesn’t remove friction — it can just scale confusion faster. What’s worked best for me: Start by running the process manually long enough to learn the edge cases. Then automate the boring, repeatable parts with clear rules and a human fallback for anything ambiguous. A simple test before automating anything: If you can’t describe the workflow in 5–7 steps (including “what happens when it goes wrong”), you’re not ready to automate it yet. Curious: what’s one workflow you automated that actually stuck long-term — and what made it stable?