I was on vacation so I didn't really post for a while!
Now I wanted to play around with Gemini Omni via the Gemini App.
Here is a small example with the "Anime" style chosen from the app with my own image and prompt
English
Dr. Fake Ink
412 posts
Dr. Fake Ink
@fakeInkDr
AI Artist | Building with Claude | building video games as a hobby | Developing a platform to customize AI image generation like never before
Latent Space Katılım Ocak 2026
124 Takip Edilen37 Takipçiler
Find the hidden object | episode 04
Difficulty: Medium
Legend of the items at the top
Automation score: 7/10 - The model kept replacing the last object with another one or covering it with something else. Upscalers also changed a few details that I needed to tune down!
Models used: GPT Image 2, NB2, Luma Uni 1.1
Original idea by @GlennHasABeard
English
Find the hidden object | episode 03
Difficulty: Easy
Automation score: 7.5/10 - had to retry a bunch of times and adjust the prompts and the upscale resolution to get to something that looks ok
**Legend of items at the top**
Idea credit goes to @GlennHasABeard
Interestingly doing this one object at the time with minimal spatial instructions to the model results with the model overriding the previously added object with the new one! I guess it is that the object doesn't belong in the image's "original" structure
English
Find the hidden object | episode 02
Difficulty: Easy
Legend of items at the top
This one needed a few generations of each image until a "suitable" image was found. So automation score is 8/10
Original idea by @GlennHasABeard
English
Dr. Fake Ink@fakeInkDr
Trying to replicate @GlennHasABeard's work, completely automated on @magnific spaces. This was fun to do and difficult as well. I found adding item after another to be better than adding them at once even with the best models. Multiple upscales and fixes are needed but in the end I managed to do the entire thing fully automated - tbh except adding the legend on the top right Should I do this more - obviously the credit goes to @GlennHasABeard for his idea. If I see some engagement to this I might do this more often and maybe drop one every day Models used: - NB2 - GPT Image 2 High - Luma Uni 1.1 High - Magnific creative upscaler - a couple of LLMs for prompting
QME
@GlennHasABeard @magnific would love to see what you come up with if you go for a fully agentic workflow
English
@fakeInkDr @magnific Really great. I haven't tried doing it with an agent workflow yet. Keep creating them!
English
Trying to replicate @GlennHasABeard's work, completely automated on @magnific spaces. This was fun to do and difficult as well.
I found adding item after another to be better than adding them at once even with the best models.
Multiple upscales and fixes are needed but in the end I managed to do the entire thing fully automated - tbh except adding the legend on the top right
Should I do this more - obviously the credit goes to @GlennHasABeard for his idea. If I see some engagement to this I might do this more often and maybe drop one every day
Models used:
- NB2
- GPT Image 2 High
- Luma Uni 1.1 High
- Magnific creative upscaler
- a couple of LLMs for prompting
English
@HappyyPablo This is so cool! I have been wanting something like this for a little app I'm working on. Awesome!
English
🤗 Model: huggingface.co/NemoStation/Ma…
🎮 Hosted Demo: vlm.nemostation.com
Training recipe + a new dense-captioning/grounding benchmark dropping soon!!
English
open sourcing Marlin-2B 🐟
a tiny VLM to extract structured information from videos
Marlin is finetuned for two questions devs want to ask in their videos: what is happening, and when?
Best open model in its weight class, competitive with Gemini-2.5-flash at only 2B params 🧵
English
This is how the space is looking right now! not that fancy but there is a lot that can be done to make this approach better
English
Deities series | episode 34
African: Olorun
Creator of all energy
English
Here’s another one!
Saturo Gojo was really a pain to do!!
Due to censoring the models, a lot of things had to be done with older/dumber models.
The upside I just decided to make a script that picks two frames that are similar from the different ending of a video and then convert them into a GIF
GIF
English
@deekaymotion In case you wondering about the process x.com/fakeInkDr/stat…
Dr. Fake Ink@fakeInkDr
Took inspiration from @deekaymotion let's walk line up, then used Gen AI to create a similar style animation. I've always respected the amount of effort and talent that was put into this, but trying to do made me respect it even more. What was used: - Luma-Uni 1.1 High for character concept - GPT 2 for sprite sheet - Seedance 2 Fast for the animation - Had to remove some frames myself - Upscaled via Magnific Creative - Then converted to GIF
English
Another attempt at @deekaymotion's Let's walk
Getting the animation to loop correctly is still such a headache
GIF
English
@dmitryrechkin @levelsio Tbh I think it is a matter of getting used to it! I mainly use CLI but I’ve been trying the apps a bit to get a feeling but nothing impressed me just yet
English
@fakeInkDr @levelsio Claude code is still a bit better but codex app is damn good
English
NVIDIA just dropped another cool world model - SANA-WM.
You give it one image and a camera path, and it spits out a full minute of 720p on a single GPU.
> handles a full 60s
> 1-minute clip in just 34 seconds (on a 5090).
> follows precise 6-DoF paths, movement feels intentional and grounded.
> Based on SANA 2.6B
World generator with cinematic fly-throughs for any fantasy setting
nvlabs.github.io/Sana/WM/
English
Took inspiration from @deekaymotion let's walk line up, then used Gen AI to create a similar style animation. I've always respected the amount of effort and talent that was put into this, but trying to do made me respect it even more.
What was used:
- Luma-Uni 1.1 High for character concept
- GPT 2 for sprite sheet
- Seedance 2 Fast for the animation
- Had to remove some frames myself
- Upscaled via Magnific Creative
- Then converted to GIF
GIF
English