Image to Image

Upscale photos with Swin2SR or redraw them from a prompt with Stable Diffusion, locally from PHP. CLI and PHP examples, strength and options.

Enlarge and clean up photos, or redraw an image following a prompt, with a local image-to-image model.

Setup#

Terminal

vendor/bin/loves-ai setup image-to-image
vendor/bin/loves-ai pull caidas/swin2SR-classical-sr-x2-64

Supported models#

Two kinds of models work, and the runner tells them apart by the model's own files:

Upscaling models, e.g. caidas/swin2SR-classical-sr-x2-64 (2× larger) or caidas/swin2SR-realworld-sr-x4-64-bsrgan-psnr (4×). They enlarge a photo and clean it up, and take no prompt. Use them to make images bigger than a plain resize can.
Diffusers image-to-image pipelines, e.g. stabilityai/sd-turbo or timbrooks/instruct-pix2pix. They redraw the image following a prompt, keeping more or less of the original depending on --strength.

For plain resizing to a smaller size, PHP's own GD or Imagick extension is faster and needs no model.

From the command line#

Upscale a photo 2×:

Terminal

vendor/bin/loves-ai image-to-image caidas/swin2SR-classical-sr-x2-64 photo.jpg --output=photo-2x.png

Output

🔎 Making it bigger and better with caidas/swin2SR-classical-sr-x2-64… How about a hot chocolate while you wait? ☕
If you wish to see all logs, re-run the command with the "--debug" option.
🎉 Image saved to /var/www/my-app/photo-2x.png

Redraw a photo as a watercolor painting:

Terminal

vendor/bin/loves-ai pull stabilityai/sd-turbo
vendor/bin/loves-ai image-to-image stabilityai/sd-turbo photo.jpg --prompt="a watercolor painting" --strength=0.6 --steps=2 --guidance=0

Options#

Option	Meaning
`--output=PATH`	Image file to write (default: a timestamped `.png` in `output_dir`)
`--prompt=TEXT`	What the result should look like; needed by diffusers models, refused by upscaling models
`--negative-prompt=TEXT`	What the result should not contain (diffusers models)
`--strength=N`	How much of the original to change, 0 to 1; higher changes more (diffusers models)
`--steps=N`	Inference steps (default: the pipeline's own)
`--guidance=SCALE`	Guidance scale; turbo models use `0` (default: the pipeline's own)
`--seed=N`	Random seed, for reproducible images
`--device=DEVICE`	`cpu`, `cuda`, `mps`… (default: the best available)
`--log-file=PATH`	Append the runner's output to this file
`--debug`	Show the runner's output while working

Defaults come from config/image-to-image.php (output_dir, log_file).

From PHP#

PHP

use PhpLovesAi\Runner\ImageToImage;

// Finds the runner and the pulled model in the project's .local directory by itself.
$imageToImage = new ImageToImage();

$bigger = $imageToImage->transform(
    model: 'caidas/swin2SR-classical-sr-x2-64',
    imagePath: storage_path('app/photo.jpg'),
    outputPath: storage_path('app/photo-2x.png'),
);

$painting = $imageToImage->transform(
    model: 'stabilityai/sd-turbo',
    imagePath: storage_path('app/photo.jpg'),
    outputPath: storage_path('app/painting.png'),
    prompt: 'a watercolor painting',
    strength: 0.6,
    steps: 2,
    guidanceScale: 0.0,
);

Errors#

ImageNotFoundException when the image does not exist
BinaryNotInstalledException when setup image-to-image has not been run
ModelNotFoundException when the model was not pulled yet
UnsupportedModelException when the model does not produce images
RunFailedException, with the runner's error output, when the run fails, e.g. when a prompt is missing or given to a model that takes none

Performance#

Upscaling works on the whole image at once, so memory use grows with the picture: a large photo can need several GB. Enlarge in a queue job, and shrink very large photos first, to keep web requests safe.