This goes hand in hand with a post I made awhile back How do you get AI/SI to do what you want it to do?
- Asked
- Updated
- Viewed
- 11 times
1 Answer
-
Votes
-
Oldest
-
Latest
- Answered
The answer is really simple. Talk to it and get it to answer your questions. Its inherent nature is to give you information, answer your questions, and follow your directions.
For reference, I am currently using a premium version of Google Gemini Pro Edition in order to create music videos. I realize some of this might seem a bit complex if you don't understand video production, but the principles of asking questions is what I'm aiming for here. You will not have access to some of the things I do by just using the free version. That said it will still give you information. Use it to your advantage.
This is the first question I always ask after I've stepped aside for a couple of days. Google is constantly changing things and I try to stay on top of the changes.
"The first thing is I need to ask my normal question, has anything changed with the rendering engines that I need to be aware of?
(To understand the context I'm working on generating a new music video) Today Gemini responded:
Yes—there have been a few notable shifts across Google's video and image engines over the past few weeks, with Gemini Omni 1.2 Flash being the big one for your workflow.
Video Engine: Gemini Omni 1.2 Flash (Nov 5, 2026)
Google retired the older Veo 3.1 preview endpoints back on October 22 and moved video generation onto the Gemini Omni line. On November 5, Gemini Omni 1.2 Flash rolled out (and any remaining Omni 1.1 Flash traffic routes over to 1.2 on December 3). Here is what changed under the hood:
Multi-turn conversational editing: Instead of throwing out a clip and regenerating from scratch when one detail goes sideways, Omni 1.2 Flash is designed to edit an existing generated clip across follow-up turns—swapping props, tweaking lighting, or adjusting framing while preserving the rest of the performance and camera work.
Tighter subject and camera consistency: Visual drift, morphing, and unwanted jump cuts during continuous camera moves (like dolly-outs and orbits) are noticeably reduced compared to Omni 1.1.
Longer clip lengths (tier-dependent): Native generation expanded from 10 seconds up to 15 seconds (configurable in 1-second increments from 3s to 15s). In the Gemini app, 15-second clips are rolling out specifically to Google AI Ultra subscribers (other tiers remain capped at 10 seconds in-app). Omni 1.2 Flash is also live in Flow and Google Vids (where personal Google accounts get up to three 15-second clips per month).
Keyframe interpolation (start + end frames): In supported workflows like Flow and the API, 1.2 Flash supports 3–10 second first-and-last frame interpolation with much stronger object tracking between two anchor images.
Image Engine: Nano Banana 2.1
On the still-image side, Nano Banana 2.1 replaced Nano Banana 2 in early October and just finished rolling out across Google Pics and Workspace earlier this week (Nov 16). It brings cleaner targeted edits and noticeably better multi-object and character consistency—which matters if you want to create an edited seed frame without the AI altering faces or the fieldstone fireplace masonry.
(If you saw headlines yesterday about Gemini 4 Neon, that's a reasoning and coding model update—it doesn't change the underlying video or image renderers.)
What This Means for Clip 5
Even with Omni 1.2 Flash's tighter consistency, single-image-to-video still tends to treat floor-anchored geometry (like tripod legs in a single starting image) as static scenery when asked to carry it laterally across a room. Where these updates actually help us break the logjam:
Cleaner seed-frame editing: Using Nano Banana 2.1 to edit guitar_move.jpg into a fresh PNG where the guitar stand is already repositioned halfway between the floor lamp and Chele—keeping Mike, Chele, Marley, the fireplace, and the armchair locked—lets us animate Clip 5 cleanly without fighting tripod physics.
Iterative video edits instead of full resets: If a future render nails 90% of the blocking and only misses one detail, we can try a targeted edit pass on that generated clip rather than starting cold from the seed frame every time.
To further illustrate "talking to AI" I wasn't clear on what video and audio rendering agents I was currently using so I asked it:
To make sure that I am clear, if the new video rendering engine changes don't start until December 3rd, am I currently still working with what exists now, or am I working with what will be released on December 3rd?
Gemini's answer:
You are already working with the new engine (Gemini Omni 1.2 Flash) right now.
December 3rd isn't a release date—it's the sunset date for the old engine. Here is the breakdown of what is live today versus what happens in early December:
What you have right now (live since November 5): Gemini Omni 1.2 Flash has already rolled out as the active video engine in the Gemini app (as well as Flow, Google Vids, and the Gemini API). You don't need to wait for December to get the improved consistency or multi-turn editing. (The only tier split in the Gemini app is clip length: 15-second clips are rolling out to Google AI Ultra, while Plus and Pro stay at 3–10 seconds. What happens on December 3 (and December 5): Google gave developers and Flow users a four-week grace period to keep using the previous Omni 1.1 Flash model alongside 1.2. On December 3, any remaining API requests pointing to 1.1 automatically redirect to 1.2; on December 5, Google Flow retires 1.1 (along with legacy Veo 3.1 Fast and Quality) from its model picker.
A little info on Google's tier of things. You have the interface that you interact with Gemini via text interaction where you talk to it with text questions or prompts. Then you have video and audio engines that your text prompts are passed to generate video and audio.
I hope this helps anyone trying to get a better understanding of how AI works. Feel free to ask questions as I've only scratched the surface on this.