4 ms·
Any idea why they didn't use GPT-4o image generation?
by cubefox 1y ago
Any idea why they didn't use GPT-4o image generation?
- Alifatisk 1y agoI think it's because Geminis nano banano is better than 4o imagegen at creating and editing images from instructions
- Jackson__ 1y agoBecause GPT-4o is too orange (literally).
- neom 1y agoSomeone who works in AI told me they think that was trained in as a "watermark", apparently the same is true with the em-dashes, to "ease people into AI" or something.
- svantana 1y agoGemini is better. See results of double-blind evaluation here: https://lmarena.ai/leaderboard/image-edit https://lmarena.ai/leaderboard/image-edit
- pwython 1y agoValid question, as they already have a partnership with OpenAI to use ChatGPT in Siri. I personally use GPT for illustrations and Nano Banana for photo edits (Midjourney for realistic photos). As an aside, perhaps they're using GPT/Codex for coding. Did anyone else notice the use of emojis and → in their code?
- whywhywhywhy 1y ago4o generates new images that try to be close but not exact it doesn't edit existing ones so wouldn't be much use for this.
- exadeci 1y agoSomeone here shared an interesting comparison of models performances, and Gemini does seem much better https://genai-showdown.specr.net/image-editing https://genai-showdown.specr.net/image-editing