|
Anna Kendrick embedding created by Airathias Main V1 embedding was created in 800 steps, and a pretty high number of vectors per token. Try to keep your prompt short, though I've seen it work fine with longer prompts. If you find the face not defined enough, just (annakendrick:1.2) it. As you'll see, it doesn't like the base model SD 1.5 much, and it leans toward better results in photorealistic renders. It does perform for certain fantasy semi-realistic artstyles, though. Check the X/Y grid linked below for a comparison with a few checkpoints. This time around I decided to offer the option of more epochs, just in case. I've created a bunch of X/Y grids for you to pick and choose: For the initial version I chose the 3 I liked best, 800, 1000 and 1500. If you'd like another epoch count embedding let me know. I added prompts to the example images, should work fine. These prompts were basically blatantly copied from other examples throughout the website. If I copied yours, mea culpa. You can use this without crediting me. The embedding was trained on stock SD 1.5, so it should work for many if not most models based off of 1.5. Images listed are made using a handful of models to test, including: Important for the examples: In the example images I use a keyword like "annakendrick-sd-v2-800", this was just for testing purposes but I've elected to keep it in there so you can see what version I ended up using for that render. In your prompt you should use "annakendrick", "annakendrick-1000" or "annakendrick-1500". Also, the model names in the examples are like "SD_15_Photoreal_NSFW_<name>". This is my local model name formatting templated like "<basemodel>_<artstyle>_<sfw/nsfw>_<name>". Hashes are shown as well, but just in case you were wondering. | 
发表于 2023-6-7 22:00:43