Tonight’s session has been a blast! We’ve officially upgraded the entire image generation system, elevating our Memory Painter to a whole new level of power and flexibility.
One of the major additions is real-time editing. For example, a friend gave me a simple prompt, and the AI generated this based on their input:
Right after, they requested: “Make the hair black, the fur black, and add a silver necklace.” The AI responded instantly with this revised image:
We’ve also introduced a feature I’m especially proud of: Image Memory — a robust image history management system.
Here’s how it works:
Reverting Back:
You can issue natural language commands like:
- “Undo last change”
- “Go back to previous version”
- “Revert changes”
- “Undo”
- “Go back one step”
Moving Forward:
If you’ve reverted and want to step forward again:
- “Redo last edit”
- “Go forward”
- “Next version”
- “Redo”
How It Works:
- Every edit is logged into a visual history.
- You can move backward and forward through each version.
- Full edit history is maintained until a new image session begins.
- Multiple-step reverts are supported: e.g., “Go back three versions.”
Feedback:
The system intelligently notifies you when:
- No prior versions exist (“Cannot revert further”).
- You’ve reached the latest version (“No more recent versions available”).
- No edit history is available for the current image.
While Memory Painter supports seamless voice and text-based control, the memory it uses is temporary — if you want to preserve your creation, you need to save it. Currently, we only retain the original text prompts in long-term memory. Visuals themselves are not stored — yet.
That said, future updates may include an extended memory module capable of capturing the BGR structure of generated images for training, regeneration, or even AI-driven stylistic recall. Stay tuned. We’re just getting warmed up.
It got most of it correct but noticed the my backend did not include the breeds so we did a correction it regenerated.
Last I asked for all of them to have party hats
The system has ability to use manual edit where you can type in the changes either directly in the image window, or in the normal message system if the image is up. You can also use your voice and just tell it to edit the and add things. The undo etc allows you to go back and forward between generations allowing you to textual or vocally photo edit.
The only downside is speed currently of image generation is slow, but like all things in time they will add more GPU’s etc to speed up this for everyone and one day it will all be instant.
Pretty cool stuff. I do also have an option although its on implements to allow masking on the image window where you can mark just like openai and other paint programs what you want to change but that will be added later as it was designed for the older model and not the new ones.
I also wanted to mention after using cursor ai for the last 24h just wow, best ai editor system I have used yet. Love it better than Visual code studio’s co-pilot. I may have to test for a month windsurf as well to see which is better. But I have been moving even faster with this tool for coding than the back and forth of openai chatgpt agents I was using. but hey that is only based on one day. Ps. Also almost capped on my usage for the month already in 1 day on the pro plan lol.
