Dataset tools — images & text

Camera names such as IMG_20260914_101501.png mean nothing to a training tool. The Renamer copies the files to Subject-Concept-0001.png, Subject-Concept-0002.png… in a random order.

1. 1
   Source Folder — the folder to rename. Sub-folders are included.
2. 2
   Output Folder — filled in for you: renamed inside the source.
3. 3
   Subject Name — the first part of the new names, for example the person or object.
4. 4
   Concept Name — optional second part: a style, a place, a tag.
5. 5
   Randomize order — shuffles the numbering so the training does not learn the original order.
6. 6
   Copy files — on by default: the originals stay untouched. Untick it to move the files instead.

Start Number and Digits set the counter (4 digits = 0001). Images only skips non-image files. Click Run.

1. 1
   Run — renames (copies) every file of the source.
2. 2
   Log — one line per file: old name → new name.

Training tools read one .txt caption next to each image. Captions creates them all at once.

1. 1
   Image Folder — the dataset (the files directly in it).
2. 2
   Fill all captions with same text — writes the text below into every file; unticked, the files are empty.
3. 3
   Caption text — for example a trigger phrase shared by the whole dataset.
4. 4
   Backup existing .txt files — an existing caption is renamed with a timestamp before being replaced.
5. 5
   Run — creates the caption files.
6. 6
   Log — one line per file created.

Remove or replace a word in every caption of a folder. Here, the tag watermark is removed from 8 files.

1. 1
   Search For — the tag, word or phrase to find.
2. 2
   Replace With — the new text. Leave it empty to delete.
3. 3
   Whole word — matches cat but not category. Case sensitive and Regex are also available.
4. 4
   Cleanup leftover commas — after a delete, removes the empty , , left in tag lists.
5. 5
   Preview Matches — counts the matches without changing anything.
6. 6
   Counters — how many files, and how many matches in how many files.

1. 1
   Run — applies the change to every matching file.
2. 2
   Status — the number of edits made.
3. 3
   Log — one line per file changed.

Collects every caption of a folder into a single prompts file, one line per caption — handy to test a model with the real captions of a dataset.

1. 1
   Output File — where the combined file is written.
2. 2
   Prefix — optional text added before every line (a Suffix is added after).
3. 3
   Run — builds the file.
4. 4
   Status — how many prompts were collected.

Resize a whole folder to a target size, by its longest or shortest side, in PNG at maximum quality.

1. 1
   Load Images — shows the images of the source folder as a gallery.
2. 2
   Target Size — in pixels, for example 1024.
3. 3
   Resize longest / shortest side — which side gets the target size; the other one follows the proportions.
4. 4
   Output Folder — filled in for you: resized inside the source.
5. 5
   Resize Selected — resizes the ticked images (Select All / Deselect All above).
6. 6
   Thumbnail — untick an image to leave it out.

Load two images and drag the slider to see the difference: an upscale, an edit, two model outputs.

1. 1
   Browse A — the before image — or drop a file on the left half.
2. 2
   Browse B — the after image — or drop a file on the right half.
3. 3
   Slider — click or drag anywhere in the view to move it.
4. 4
   Swap A/B, Fit, Clear — exchange the two images, reset the zoom, empty both slots.