Main Window
Style image – Landscape from Saint-Rémy / Vincent van Gogh / Public domain
1. New project
Starts a new project.
2. Open project
Opens a saved project.
3. Save project
Saves the current project. All content/style images opened, saved selections and style maps belong to them, seed value(for text-to-image) and input text will be stored in the project file.
4. Save project as
Saves the current project to a different project file.
5. Undo
Undo the previous action.
6. Redo
Redo the previously undo-ed action.
7. Style scale
Configures the maximum scale of style image patterns to be used in both prototyping and finalization processes.
A swirl in “The Starry Night” is a good example of a coarse scale pattern, while individual brushstrokes constituting it are fine ones.
In general, as coarser scale patterns are being used, the original “content” of the content image is getting more vague and vice versa.
8. Smoothing level
Configures the level of the smoothing effect at the boundary of a style in the prototyping process.
When painting the prototype over the content image, it slightly goes beyond the boundary of the target region for smoothness, and this parameter sets how farther it does so.
Note that it’s only used to make a prototype and not used in finalization.
9. Prototyping
Generates a prototype image, which contains the “content” of the currently selected region in the content image while having certain spatial scale image patterns of its counterpart in the style image, in an interactive manner(takes a few seconds), and paint it over the content image. When painting the prototype, it slightly goes beyond the boundary of the target region for smoothness. (See Smoothing level)
And more importantly, at the same time it stores this spatial style configuration in the style map, which will be used in the finalization process. To start finalization, there should not be any region not mapped to a certain style(which is identified by 3 parameters – a style image, a region in that style image and a style scale). You can also use multiple style images to stylize a content image, although you must be aware that as the number of distinct styles increases(especially using more than 1 style image), so do memory usage and execution time.
Sometimes prototype images are just good enough that the finalization step is not needed at all, but in general, prototype images only act as starting points of finalization and give some hints to users about what final results would look like. In addition, in the finalization step, users can ignore it and choose to start from the original content image if they want.
10. Erase style
Erases the painted prototype from the current selection of the content image, and also updates the style map correspondingly.
11. Finalization
Executes the pop-up dialog which will run an iterative algorithm to stylize the image using the configured style map and the prototype.
12. Enable/Disable seed input
“Seed” here means a non-negative integer used to initialize a pseudorandom number generator, whose number sequence is completely determined by it.
Since text-to-image generation uses random noises, if users want to reproduce a certain image(or sequence of images), the seed value used when it was generated must be provided as well as the image size, the number of steps and input text.
- If unchecked(default) : the seed input edit box remains deactivated and shows the seed value used in the last(or ongoing) text-to-image run, which will be set to a random value each time.
- If checked : the seed input edit box gets activated so that users can edit its value which will be used as the seed.
During a Text-to-Image run, seed will be incremented by 1(if seed is equal to the maximum value 18,446,744,073,709,551,615 then it’s set to 0) after generating each image for the current line of text, and will be reset to the initial value after finishing each line of text.
Let’s see an example. An user sets the number of images(per line of text) to 5, provides 2 lines of text and starts Text-to-Image, without enabling seed input. At first, the initial seed will be auto-generated(let’s say 7) and displayed in seed input edit box. Then 5 images will be generated from the first line of text, with seed value 7,8,9,10 and 11 respectively, followed by another 5 images for the second line with the same sequence of seeds. And each of those images can be reproduced later with the corresponding seed value and the line of text.
13. Seed input
Please refer to the above section “12. Enable/Disable seed input”.
14. Width of images
Width of images to be generated from the input text. Default value is 512 or 768, according to the type of diffusion model(SD 2.0-base or SD 2.0-v).
15. Height of images
Height of images to be generated from the input text. Default value is 512 or 768, according to the type of diffusion model(SD 2.0-base or SD 2.0-v).
16. Number of images
The number of images to be generated from each line of the input text.
17. Number of steps
The number of diffusion steps when generating images from the input text. You may adjust the balance between execution speed and image quality using this parameter.
18. Input text for image generation
Text describing images to be generated. Each line will be processed individually.
19. Open text editor
Opens the text editor dialog so that it can be edited more conveniently.
20. Text-to-Image
Generates images from the input text, using Stable Diffusion 2.0.
(Please note that Teripic uses slightly modified version of it so that result images will not be exactly same as ones from the original.)
21. Weight
Proportion of diffusion steps to be run when generating images from input text and an initial image, compared to usual Text-to-Image.
Higher values result in images less similar to the initial one and vice versa.
22. Text-to-Image with an initial image
Generate images from the input text and the current content image, using Stable Diffusion 2.0.
(Please note that Teripic uses slightly modified version of it so that result images will not be exactly same as ones from the original.)
Compared to usual Text-to-Image which starts from a completely random noise image, it starts from a noised version of the initial image.
23. Open image
Opens a image.
24. Export image
Exports the current image.
25. Zoom in
Zooms in on the current image.
26. Zoom out
Zooms out on the current image.
27. Zoom scale
Shows the zoom scale of the current image.
28. 2x Upscale
Upscales the current image to 2x size.
29. 4x Upscale
Upscales the current image to 4x size.
30. Scale image
Scales the current image to an arbitrary size.
31. Rotate left
Rotate the current image by 90 degrees in a counterclockwise direction. The current selection, saved selections and the style map will be also rotated accordingly.
32. Rotate right
Rotate the current image by 90 degrees in a clockwise direction. The current selection, saved selections and the style map will be also rotated accordingly.
33. Expand/Collapse selection view
Expand/Collapse selection view, which contains current selection view, saved selection tab and style map tab.
34. Free select
Clears the current selection and enters “free select” mode, which means it’s ready to begin selecting a region in the image. You can draw arbitrarily shaped boundaries by click-and-dragging as well as polygonal boundaries by clicking vertices, and you can use both methods alternately.
35. Free select plus
Enters “Free select plus” mode(without clearing the current selection), which means it’s ready to begin selecting a region in the image to be added to the current selection when finished. Also, if you click a saved selection in this mode, it is added to the current selection.
36. Free select minus
Similar to “Free select plus” except it does subtraction instead of addition.
37. Select all
Selects the whole image region.
38. Save selection
Saves the current selection. It can be retrieved in “Saved Selection” tab.
39. Delete selection
Deletes a saved selection which is active(highlighted) in “Saved Selection” tab.
40. Import selection
Imports saved selections from a Teripic selections file(*.trs).
41. Export selection
Exports saved selections to a Teripic selections file(*.trs).
42. Maximize/Unmaximize window
Maximize/Unmaximize this window it it is undocked(floating). This icon is disabled when it’s docked.
43. Dock/Undock window
Dock/Undock this window.
44. Current Selection
Shows the currently selected region of the image(white for the selected and black otherwise).
45. Saved Selection Tab
Shows saved selections.
If you click a saved selection,
- in “free select” mode, it will be loaded to the current selection.
- in “free select plus” mode, it will be added to the current selection.
- in “free select minus” mode, it will be subtracted from the current selection.
46. Style Map Tab (Only exists in the content subwindow)
Shows distinct stylized regions in the content image which are categorized by stylization parameters – a style image, a region in that style image and a style scale. Note that this is not related to saved selections at all, as you can apply a style without saving any selection.
If you click an item of the style map tab, the corresponding content/style image regions will be loaded to the current selection of each content/style window and the value of the “Scale” combobox will be changed to the corresponding style scale.
47. Image Tab
Shows the list of the images opened. You can change the order of the opened images by drag-and-dropping tabs.
48. Image Page
Shows the current image and the current selection boundary. You can select the certain region of the image by clicking or dragging on it.
Finalization Dialog
1. Zoom in
Zooms in on the current image.
2. Zoom out
Zooms out on the current image.
3. Zoom scale
Shows the zoom scale of the current image.
4. Content scale
Configures the scale of the “content” to be preserved. If too much fine-scale details of the content image that has to be preserved disappeared during finalization, you can lower this scale and try again, and vice versa.
Values from 3 to 5 are generally recommended.
5. Content weight
Configures the relative importance of “content” compared to “style”. This affects finalization much more directly than the content scale parameter. For example if you set this to 100% and run finalization from the original content image(not the prototype), it will be not altered at all, because that 100% dictates “content” must be perfectly preserved while puts zero importance on “style” and the original image just trivially satisfies this condition.
Values from 5 to 20 percent are generally recommended.
6. Iteration
Configures the number of iterations to run the finalization algorithm. In general 2000 iterations when starting from the original(1000 if from the prototype) are enough to get a best quality result. You can also incrementally run it (first 500~1000 iterations to see what it looks like and another 1000~2000 to complete).
7. Use prototype
Chooses whether to use the prototype as the starting point or not(in this case the original content image will be used).
8. Time remaining
Shows estimated time remaining until the current run finishes.
9. Time elapsed
Shows time elapsed for the current run.
10. Finalize/Stop
Starts/Stops finalization.
11. To main
Sends the current image to the main window.
12. Export
Exports the current image. You can export it while finalization is still running.
13. Progress bar
Shows the progress of finalization.
14. Image page
Shows the current image being processed. It is updated every 3 second during finalization.
Shortcut
| Key | Task |
|---|---|
| Ctrl+N | New project |
| Ctrl+Shift+O | Open project |
| Ctrl+S | Save project |
| Delete | Erase style |
| Ctrl+Z | Undo |
| Ctrl+Y | Redo |
| Ctrl+O | Open image |
| Ctrl+A | Select all |
| Esc | Clear current selection |
| F1 | Free select |
| F2 | Free select plus |
| F3 | Free select minus |
Command Line Arguments
You can use command line arguments listed below, by directly executing “program\Teripic.exe” instead of windows shortcuts.
- -v768 : uses the Stable Diffusion 2.0-v model(trained for 768×768 resolution) instead of Stable Diffusion 2.0-base(trained for 512×512 resolution). Please check here for further information.
- -cuda <index> : selects one of NVIDIA Graphic Cards if there is more than one. Index starts from 0.
- -cpu : uses CPU instead of Graphic Card. It’s not recommended as it will be far much slower.