#Hard time generating images.

8 messages · Page 1 of 1 (latest)

jovial abyss
hollow apex
slate prawn
#

thanks for the heads up! image editing isn't yet possible

#

recommend trying prompts like this:
generate an image of a purple and blue dinosaur, waving at a four pointed star that is also purple and blue. it should be on a black background and stylistically be like an emoji.

e.g. https://g.co/gemini/share/8ae6b0b5725c -- and as you see, it's not perfect (the star has 5 points, not 4)

Gemini

Created with Gemini

#

it's generally better to try to prompt again, since gemini can't edit what it previously generated so stuff like "make it X" doesn't work as well as just reprompting -- we're working on making this possible, but it's not yet there

#

you can also get a lot out of iterating on your prompt -- e.g.

generate an image
An image of a dinosaur waving a four-pointed star.
A chubby, cartoonish dinosaur with rounded edges and smooth gradients transitioning from deep royal purple to electric blue stands centered against a pure black background, its short arms raised in a friendly wave with stubby three-fingered hands.
A luminous four-pointed star hovers in the upper right corner of the frame, its points perfectly symmetrical and sharp, featuring the same purple-to-blue gradient as the dinosaur but with an added inner glow that creates a soft halo effect against the dark backdrop.
The dinosaur's large oval eyes sparkle with white circular highlights while its broad, permanent smile curves upward in typical emoji style, and its entire form has a subtle glossy sheen as if rendered in polished plastic or smooth vector graphics.

gives me:
https://g.co/gemini/share/77fdecce9936

#

would love to hear what you'd like to see for emoji creation!

jovial abyss
# slate prawn it's generally better to try to prompt again, since gemini can't edit what it pr...

That's really interesting! I often find that asking for changes from Gemini rather than reprompting often ends up better, and less drastically different IMO. I've assumed that when Gemini changes the prompt that it flushes out what I asked for and fits it in while maintaining the overall prompt and seed and whatnot.

One thing I like that I've seen in other AI products is the human asking AI for an image, the AI writes a prompt and executes it, then the human is able to hop in and directly tweak the prompt that the AI wrote rather than relying on the AI to know what parts to change.

This is an example of asking Gemini working fairly well all things considered IMO. https://g.co/gemini/share/367006d94720

Gemini

Created with Gemini