Screen-chanTV Merge Illustrious
Steps: 15-30+
CFG scale: 3-7
Sampler: euler a or DPM++ 2M Karras
Vae Included
positive prompts:
newest,very awa,masterpiece,high quality,high resolution,amazing quality,best quality, good lighting,detailed eyes,anime coloring,anime screencapnegative prompts:
worst quality,bad quality,low quality,lowres,scan artifacts,jpeg artifacts,sketch,bad quality,jpeg,artifacts,signature,username,text,logo,bad anatomy,artist name,artist logo,extra limbs,extra digit,extra legs,extra arms,blurry background,simple background Description
The blurry backgrounds from the previous version have been fixed.
FAQ
Comments (4)
Amazing checkpoint for those that look for one able to render pics with a similar style than the Danbooru "anime screenshot" one.
One question, tho. You suggest using "newest,very awa,masterpiece,high quality,high resolution,amazing quality,best quality, good lighting,detailed eyes,anime coloring,anime screencap" as a positive prompt. Many of those "tags" aren't Danbooru ones. For exemple most of your suggested "tags" are a relic inherited from the past either on the original Illustrious ckeckpoint or on trained ones using custom tags, and no one today is certain about their actual effects on generated images (masterpice, high quality, best quality, amazing quality. newest, detailed eyes, etc...).
So could you elaborate regarding your positive and negative prompt suggestion? Because i suspect that all of those non Danbooru tags are useless and make no actual change because of them not being interpreted by the text encoder.
I m gonna test a couple renderings using your positive prompt and using only "anime coloring" for as far as this is the only Danbooru tag among those in your suggested positive prompt and check if there is a visible difference other than the one coming from the prompt token weight change.
I understand that they are groupings, like highres, or absurdres that you can find as a tag on Danbooru, in which all the images that have that tag look aesthetically better, as for very awa it is almost the same
@CuervoTensor So i just tested on a couple pics and as i said before the positive prompt makes no difference at all. This is the same for the negative prompt when not using Danbooru tags. Also regarding actual Danbooru tags i can"t help but think that you don't understand their use. For exemple the "jpeg" tag is a tag that is related to the filetype. It makes no sense to use it. Not to mention that this is a deprecated tag. Meaning that it isn't used anymore on Danbooru.
So Here are 2 pics generated using the same parameters to test the impact of both the positive and negative prompt.
1st picture (positive and negative prompt included): https://ibb.co/Q7p3VQdD
2nd picture (no postive/negative prompt): https://ibb.co/1Gw5Tv6F
No visible difference that you could relate to tags when they aren't Danbooru tags.
So my advice would be to remove the positive prompt recommendation and remove all non-Danbooru tags or useless tags from the negative prompt (or remove the negative prompt completely and add tags to it on a per case basis).
It says commercial use is not allowed but the license says otherwise.