image generator not good related to story\character description

[azhestov]

Is a possible add to auto prompt - parts of description like appearance, body, age, features, so on?

Is a hard to parsing by code, but we have a text model here which able to parse scenario to prompt keywords at loading and place it to autoadd for any prompt start. (may be not a bad idea to have a option - possibility to convert reply to set of sd tokens related to image, before generate it. it can eat i bit of time, but is can be nice for users who want it.)

(and by the way - cut long comment to last part is a not good deal for prompt, don't be scared for long prompts for SD models - it's works)

[LostRuins]

You can add a "style" to your prompts which will be inserted into every image

For the autogenerate, it will use approximately 400 characters of length. It needs to be truncated as otherwise all the images will look the same. I will increase this a bit.

[azhestov]

Well... I used to common way when style token written last and have a little effect on the image content but work well with the hole style of the finished composition. Okay, I'll try.

About auto generation, yes, images can look the same for same reply especially if you not randomize seed every time(hint - randomize cfg scale in range 5-8(or +-1 from settings) can be a good solution too). But for ai replies most often action\situation writed first and feeling last, so if you take last part of reply for generation it is difficult to obtain an image that matches the situation. (i try make a prompt for filtering reply of different models and convert it to useful tokens but so far the result is not enough close to desired)

P.S. Best default for Euler\Euler A - is a 22 step with cfg 7, not 20.