I don't think it's anything to do with the underlying model, it is an extra step that they do to process the image with a background remover model, which looks like they haven't prioritised just now. The model itself can't just generate transparent pixels
MULTIMODAL AI
-
GPT Image 2 Proves Valuable for Application Development
By
–
I’ve been very pleasantly surprised by how useful GPT Image 2 is for app building:
-
Gemini Document Creation Falls Short of Frontier Standards
By
–
Gemini now can create documents, and it is a nice start, but not up to the frontier yet, as you can see from my "LBO of Hogwarts" test. PowerPoints are substantially worse than NotebookLM, spreadsheets are primitive, still no thinking trace, it doesn't think hard enough, either.
-
Video Generation AI Offers New Control and Execution Efficiency
By
–
Cuando hace unos años se decía que las IAs de generación de video nos iban a dar una versatilidad y control enorme nos referíamos a cosas como estas… y a más que está por venir!
— Carlos Santana (@DotCSV) 29 avril 2026
El control sumado a la velocidad de ejecución y eficiencia que la IA ofrece no tiene competencia. https://t.co/lkJrYF9H23When a few years ago it was said that video generation AIs were going to give us enormous versatility and control, we were referring to things like these… and to even more that's yet to come! The control combined with the execution speed and efficiency that AI offers has no
-
Stateful memory in InVideo Agent One bridges session and shot continuity
By
–
on parle beaucoup de génération
— Jouhatsu | AI Influence Operator (@Jouhatsu_ai) 29 avril 2026
mais le vrai sujet, c’est la mémoire
avec InVideo Agent One :
→ chaque plan n’est plus isolé
→ chaque session n’est plus un reset
→ chaque ajustement n’est pas à refaire
tout reste connecté
et ça, c’est un énorme gap https://t.co/NUbRKoVAnjon parle beaucoup de génération mais le vrai sujet, c’est la mémoire avec InVideo Agent One : → chaque plan n’est plus isolé
→ chaque session n’est plus un reset
→ chaque ajustement n’est pas à refaire tout reste connecté et ça, c’est un énorme gap -
GPT-5.5 and GPT-Image-2 Combo for Web App Development
By
–
GPT-5.5 + GPT-Image-2 is becoming one of the best combos for building apps! @dkundel breaks down why it works so well. We built those learnings into the Build Web Apps plugin, so Codex can handle the design-to-app loop for you.
-
AI Agents for Visual Demonstrations: VITUR and Even Realities
By
–
If you need your agents to be able to show you things @getVITURE or @EvenRealities are the way to go
-
Mentra Smart Glasses Limitations: No AR Without Additional Sensors
By
–
What are you looking for glasses to do? Mentra only has microphone, camera, and speaker so can’t do augmented reality yet.
-
GPT-Image-2 Used as Stop Motion Video Generator Frame by Frame
By
–
GPT-Image-2 can be a video-gen model if you want it to be!
— Matt Shumer (@mattshumer_) 29 avril 2026
Just ask ChatGPT to generate stop motion video, frame-by-frame.
It works shockingly well and is super controllable! pic.twitter.com/Q4GT6SXgj1GPT-Image-2 can be a video-gen model if you want it to be! Just ask ChatGPT to generate stop motion video, frame-by-frame. It works shockingly well and is super controllable!
-
Seedance Outperforms HappyHorse 1.0 in Subjective Benchmark Tests
By
–
HappyHorse 1.0 feels benchmaxxed… Seedance is beating it (subjectively) in every single test I’ve run so far