NHTickler41M
TMF Expert
- Joined
- Oct 27, 2018
- Messages
- 300
- Points
- 28
That is me!Unless you're scholardrb41... in which case, I already followed you yesterday!
That is me!Unless you're scholardrb41... in which case, I already followed you yesterday!
Emphasis on "sometimes."Ideogram has just introduced a new rendering model. It's sharper and higher definition... but also seems to understand some prompts less well than the previous version (which is also still available - you can choose either, and the new one uses more credits).
BUT, it seems the new model MIGHT at least sometimes understand what tickling is. I've tried some prompts specifying someone tickling someone else and have gotten some better results than before. (It still doesn't know what stocks are, and it still can't seem to handle nylons in closeup.)
I noticed this and played around with it yesterday. I seem to remember this when 2.0 came out, it looked much flashier but seemed to get the prompts confused more, however this smoothed itself out considerably over the following days and weeks. I have a feeling 3.0 will "learn" rather quickly and will be a much different animal in a week or two.... the stocks thing is infuriating, even when basing your work off of a sample pic it just does not get the concept of stocks...the nylons thing as well, ive gotten some anime style stuff that has done nylons fairly well but the rest of the styles cant seem to pull it off...Ideogram has just introduced a new rendering model. It's sharper and higher definition... but also seems to understand some prompts less well than the previous version (which is also still available - you can choose either, and the new one uses more credits).
BUT, it seems the new model MIGHT at least sometimes understand what tickling is. I've tried some prompts specifying someone tickling someone else and have gotten some better results than before. (It still doesn't know what stocks are, and it still can't seem to handle nylons in closeup.)
I noticed this and played around with it yesterday. I seem to remember this when 2.0 came out, it looked much flashier but seemed to get the prompts confused more, however this smoothed itself out considerably over the following days and weeks. I have a feeling 3.0 will "learn" rather quickly and will be a much different animal in a week or two.... the stocks thing is infuriating, even when basing your work off of a sample pic it just does not get the concept of stocks...the nylons thing as well, ive gotten some anime style stuff that has done nylons fairly well but the rest of the styles cant seem to pull it off...
I love the robot/Androidtortured in space, part whatever
I somehow got chatgpt.com to create these. But I did so by each time modifying the prompt slightlyThe AI does actually "know" what stocks are... if you upload a picture with stocks, the auto-description will sometimes refer to stocks. But it won't generate an image with stocks, even based on a source image that already has them, as you said.
I have no clue why that's the case. It's possible it's programmed not to, for some reason. Likewise the AI knows what tickling is... I've had auto-descriptions of uploaded images say "her feet are being tickled." But if you put "her feet are being tickled" in a prompt, it rarely generates that. I did get a few successes with that yesterday with 3.0, but it seems like it became less common even just in the first few hours of the new model.
Well, it's progress!I somehow got chatgpt.com to create these. But I did so by each time modifying the prompt slightly
Wan 2.1, image to video is pretty great
As far as stocks, I've had luck by simply being as descriptive as possible. For example: foot stockade pillory made of wood with two circular tight ankle holes, frame is rectangular and perpendicular to ground, woman sitting in chair behind frame with outstretched legs with bare feet through the ankle holes, soles showing.
Then, so it doesn't just give me a sort of wooden ankle cuffs device, I'll add: the rest of the front of the stockade is made of solid wood.
Do this for every element you think is important. Here's a great video on effective language https://www.youtube.com/watch?v= Od3FRMLqwFk (remove space). For example, use the word "extremely" if you want something extremely detailed, or use "beautiful" to fix deformities. Not mentioned in the video, but my personal favorites that I use frequently are "well-proportioned" and "anatomically correct" for all human characters. But be efficient, so instead of putting that behind every body part, I just say "every aspect of her body is..." etc.
Well, I kept fiddling, and this is by far the closest I've every gotten in Ideogram.
But so far I haven't been able to integrate the effect into a more complex scene, even using the same wording in the prompt.
View attachment 1066971