Base Image (img2img)

img2img = image to image

Sometimes using just a prompt isn't enough to get the image you want. Adding in a base image can help guide the AI to getting the desired results.

Variation Strength plays a large role in the amount of changes when using img2img

Settings used in examples

"text_prompts": "close up, cute bunny, tall trees, shadows, high noon, god rays",
"negative_prompts": "nude, nsfw, purple, mutated, deformed, amateur drawing, lowres, text, error, signature, watermark, username, blurry, censorship, sketch, monochrome",
"strength": 0.21,
"init_image": "...",
"width": 768,
"height": 768,
"models_dir": "C:\\models\\Artroom",
"ckpt": "Models\\aZovyaPhotorealUltra2GB_v10-fixed.safetensors",
"vae": "mse.vae.pt",
"loras": [],
"controlnet": "none",
"use_preprocessed_controlnet": false,
"remove_background": "face",
"use_removed_background": false,
"sampler": "euler_a",
"steps": 30,
"cfg_scale": 7,
"clip_skip": 1,
"seed": 82413648,
"mask_image": "...",
"invert": false,
"palette_fix": false,
"image_save_path": "C:\\ArtroomOutputs\\img2img",
"n_iter": 1,
"save_grid": false,
"speed": "High",
"long_save_path": false,
"highres_fix": true,
"show_intermediates": true,
"id": "3318777549530247"

Without any starting image this is the output image

But placing this starting image in the space provided,

These are the outputs it generates at different variation strengths.

Keeping that starting image and other settings but changing the prompt to "space ship, red nebula, large star, vibrant" will yield this result.

ControlNet Is anther method for copying details from a base image that follows certain details more accurately.