Adding to Cart…
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.You currently have no notifications.
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Well, it's certainly a leggy model - 100% leggy.
Hands are still a problem...
Regards,
Richard
I wonder why hands are so complicated to get right. Faces are OK, why are hands such a problem for AI?
If talking about AI in general, I have seen some pretty bad faces too
Humans are all different but still similar enough that anything unusual stands out. That’s why AI can mess up when generating some objects, like hands, and it’s super noticeable. You might not care if candies are in weird shapes, but fingers are another story. They’re small, detailed, and tricky to generate right, with things like finger positioning and joint movements being hard for AI to perfect.
At least, that’s how I see it.
Can you remind me which is the front of a boat, the blunt end or the sharp end? Prompt 'Spray type yacht in rough sea sailing between icebergs."
The image above has the boat going left (from the wake and towards the pointy end), but all the sails are powering it to the right, so the blunt end is the front, and this is confirmed by the bowsprit fragment visible unconnected to any other part of the vessel. Then there is the next 'Which way forward' conundrum:
When I initially tried it, I had the prompt as 'sailing through icebergs' rather than 'sailing between icebergs' and got this very literal interpretation (though at least the sails are driving it towards the pointy end. Not sure how they are so full having such a flat, unruffled sea, but there you go):
I wish the AI understood what a 'Spray Type' vessel was - one that's like Joshua Slocum's Spray, a beautiful little tub of a boat that was the first recorded vessel to go round the world single handed. It then took another 60 odd years before a solo vessel went round without stopping.
Regards
Richard
You can get a lot better results if you use a rendered model and do image2image, with text prompts. That can also help setup the "stage" for the scene, as AI otherwise pulls a random scene to start building off. It will layer and layer more details, until it has created one totally random "content scene", using text as a loose starting point, or an image as a more specific pre-selection starting point.
The big thing is that image2image will often have less finger issues, less extra appendages, rarely lose appendages, and be closer to your desire. Which you can change by changing the image you supply as the starting reference. You can even put a cheap background, painted with low effort, as a guide for the AI to use. (Dark blue water, light blue sky, white and gray clouds, green plants and grass, brown tree trunks and branches, badly drawn buildings, etc)
That was April (above) Same prompt & seed in October 2024, and the helicopter is present in the same way it was in May and July. This image has the same odd tail, odd upper lip on the figure and odd left hand as May and July:
Really feels as if no development of the AI has occurred since May.
Regards,
Richard
hands... holding hands...
Most impressive. Never seen the like before.
Richard