Summer Job
Did a new test with text and multiple characters in the same scene. Had to combine multiple techniques to get characters with different body types and readable text simultaneously. Each character was generated individually at low settings and composited together and given text to create a base image for the following iteration steps. It was ran through at a low strength a few times for cleanup, turning the cheap low settings generations into something better, fixing the hand drawn background extension, making the logos clearer, and upgrading the chalk board. Here’s what I’ve discovered. Text must be large to be preserved through generations. Low strength reference image iterations preserve text well while still cleaning up scenery. I recommend a text size around the size of Marine’s dialogue and a reference image strength of .4 which is perfect for multi-character iteration.
顯示參數
- 尺寸
- 1280 x 1280
- 取樣步數
- 19
- 取樣方式
- Euler a
- CFG Scale
- 7
- 反向提示詞
- worst quality, low quality, normal quality, bad hands, bad feet, bad anatomy, bad proportions, cropped, missing fingers, jpeg artifacts, signature, watermark, username, artist name, (short), big head, 3D, Source Filmmaker, flat color, SFM, naked_apron, child, loli, impossible_shirt
使用的模型和 LoRA
相似作品
展示更多c
Sweetie fox
Chelovek
a
Diives Artstyle, quavernsfw style, miyazero, anime coloring, solo, GyaruBulbasaur, green hair, cone hair bun, red eyes, white pupils, large pechos, thighs, Starbucks green apron, nake, holding a glass of coffee, cafeteria background, por detrás
axolpaint
fd
Marshmallow Mewgenics Tinkererdingle Style
fdfs dfddsd
a
Diives Artstyle, quavernsfw style, miyazero, anime coloring, solo, GyaruBulbasaur, green hair, cone hair bun, red eyes, white pupils, large pechos, thighs, Starbucks green apron, nake, holding a glass of coffee, cafeteria background, por detrás
axolpaint
留言3