What each verb in the rail does: what it accepts, the options in the inspector's Options tab, their defaults, and what it writes. Labels below match the UI.
Filesmith never overwrites a source or an existing file.
- A file is written next to its source as
name.<new ext>when that name is free. - If it is taken, a tag is added:
name (converted).ext, thenname (converted 2).ext, and so on. The tag is the verb:converted,compressed,resized,upscaled,no-bg,text,pages,merged. - Operations that produce many files write a new folder, such as
name (pages), withname (pages) (2)on a clash. - A failed or canceled job leaves nothing behind.
- Every verb except Generate shows a size estimate card under its options once files are queued.
Changes format. Works within a family; a file is never offered its own format. A selection never mixes two families, and the target list only shows formats valid for every selected file.
| Source | Targets | Engine |
|---|---|---|
| Image | PNG, JPG, WebP, AVIF, JXL, TIFF, BMP, GIF, ICO | ImageMagick |
| Video | MP4, MKV, MOV, WebM, AVI, GIF | ffmpeg |
| Audio | MP3, M4A, AAC, OGG, OPUS, FLAC, WAV | ffmpeg |
| Word doc, plain text, PDF | PDF, DOCX, ODT, RTF, TXT, HTML | LibreOffice |
| PDF (extra) | CBZ, CBR, CB7, CBT | mutool + 7-Zip |
| Spreadsheet (xlsx, xls, ods, csv, tsv) | PDF, XLSX, ODS, CSV | LibreOffice |
| Slides (pptx, ppt, odp) | PDF, PPTX, ODP | LibreOffice |
| Archive (zip, rar, 7z, tar, cbz, cbr, cb7, cbt) | CBZ, CBR, CB7, CBT, ZIP, RAR, 7Z, TAR, PDF | 7-Zip |
Routing notes:
- PDF to TXT uses mutool and does not need LibreOffice.
- Other document conversions use the LibreOffice copy bundled with the app. If it is missing,
Filesmith falls back to an installed LibreOffice
(
winget install TheDocumentFoundation.LibreOffice). - Archive to archive is extract and repack. Archive to PDF puts the images in filename order, the way a reader shows them.
- PDF to a comic format renders each page to an image, then packs them.
- CBR and RAR output need WinRAR (
Rar.exe), which cannot be bundled. Without it, those targets are greyed out with "WinRAR not found". Reading RAR needs nothing extra. - Images going to JPG or BMP are flattened onto white. ICO gets sizes 256 to 16.
Options:
| Option | Shown for | Choices | Default |
|---|---|---|---|
| Format | all | targets above | WebP for images, else the first valid target |
| Quality | images | smaller, balanced, best | balanced |
| Compression | archive to archive | store, normal | store |
| Resolution | PDF to comic | 72 to 400 dpi | 150 dpi |
| Page format | PDF to comic | jpg, png | jpg |
| Page quality | PDF to comic, jpg pages | 10 to 100 | 100 |
| Location | non-archive converts | next to source, chosen folder (Choose folder) | next to source |
| If file exists | non-archive converts | add (2), the only choice | add (2) |
Output: one file per source in the target format, tag converted.
Shrinks file size. Accepts images (JPG, PNG, WebP, GIF, TIFF, AVIF, JXL), video, audio and PDF. BMP, HEIC, SVG and other formats that cannot be re-encoded safely are not offered.
| Kind | Option | Choices | Default |
|---|---|---|---|
| Image | Format | Keep format, WebP, AVIF | Keep format |
| Image | Quality | 10 to 100 | 80 |
| Video | Codec | H.264, H.265, AV1 | H.264 |
| Video | Scale | 25% to original, step 5 | original |
| Video | Quality | 10 to 100 | 80 |
| Audio | Codec | Keep format, MP3, AAC, Opus | Keep format |
| Audio | Bitrate | 320k, 256k, 192k, 128k, 96k, 64k | 192k |
| Level | Lossless, High quality, Balanced, Smallest | Balanced | |
| Colour | Convert to greyscale (checkbox) | off |
Output, tag compressed:
- Image: same format via CaesiumCLT (or ImageMagick for formats Caesium cannot write), or WebP/AVIF via ImageMagick.
- Video: always
.mp4(ffmpeg). All audio tracks are kept; subtitles are dropped. Below original scale the inspector lists each file's exact output pixels. - Audio: the chosen codec's extension. Keep format on a lossless source (FLAC, WAV, AIFF) writes a max-compression FLAC.
- PDF: Lossless runs
mutool clean(no image changes). Other levels use Ghostscript, which downsamples embedded images.
Scales images with ImageMagick. Accepts images. Animated GIFs keep their frames.
| Option | Choices | Default |
|---|---|---|
| Mode | percent, dimensions | percent |
| Percent | 5% to 200% | 50% |
| Width and height | numbers; leave one blank to scale by the other | blank |
| Fit | keep aspect, stretch | keep aspect |
Keep aspect fits inside the box, so one of the two numbers may have no effect; the output size list shows the real result per file. Stretch uses both numbers.
Output: same format as the source, tag resized.
Enlarges images 2x to 4x with AI. Accepts images. Output is always PNG, tag upscaled.
| Option | Choices | Default |
|---|---|---|
| Factor | 2×, 3×, 4× | 4× |
| Model | Real-ESRGAN models found on disk, plus AI models on NVIDIA | Photo |
| AI model | imported ComfyUI upscalers, PiD (diffusion), 4× | first imported model |
| GPU mode | full GPU usage, balanced | full GPU usage |
Engines:
- Real-ESRGAN (ncnn, Vulkan). Bundled and works on any Vulkan GPU, including AMD and
Intel. The Model list is read from disk: Photo, Anime and any other shipped models, plus
your own. Add your own model opens the user models folder; drop in a
.param/.binpair and it appears, marked "added by you". - AI models (NVIDIA only). Shown when an NVIDIA GPU is detected; otherwise the inspector
says why. Opens a second picker:
- ComfyUI upscalers. Choose your ComfyUI folder once; Filesmith reads
models\upscale_modelsin place (nothing copied, ComfyUI does not need to run) and loads each file with spandrel. Each entry shows its native scale; unverified ones say "experimental". See models.md. - PiD (diffusion). Offered only when installed or when its weights can be reused from your ComfyUI. The PiD engine card downloads about 6 GB once (non-commercial use only); PiD install removes it. PiD has no GPU mode. Under 12 GB of VRAM it warns that large images are reduced to fit.
- ComfyUI upscalers. Choose your ComfyUI folder once; Filesmith reads
- GPU mode. Balanced caps VRAM and paces the work so other apps stay responsive.
The output size list shows each file's result. Outputs over about 1 GB trigger a warning.
Cuts the subject out with rembg (BiRefNet model, alpha matting and mask clean-up always on).
Accepts images. Needs the free uv tool (winget install astral-sh.uv); the first run
downloads the model once (a few hundred MB), then works offline.
| Option | Choices | Default |
|---|---|---|
| Fill | Transparent, White, Black, Green screen, Custom Color, Custom Image | Transparent |
| Custom colour | colour picker (Custom Color) | #ff0000 |
| Background image | Choose image (Custom Image) | none |
Output: PNG, tag no-bg. Custom Image composites the cutout over the chosen image.
Creates images from a text prompt with your local ComfyUI. Takes no dropped files; the prompt is typed in the centre. Filesmith attaches to a running ComfyUI or starts its own headless one. If ComfyUI is not found, a COMFYUI group asks you to locate it.
| Option | Choices | Default |
|---|---|---|
| Model | image models in your ComfyUI folder | first runnable model |
| Negative prompt | text (hidden for models that ignore it) | blurry, low quality, watermark, text |
| Style | none, realistic, photographic, anime, artsy, 3d render, fantasy | none |
| Size | Square 1:1 · 1024, Portrait 2:3, Landscape 3:2, Portrait 9:16, Landscape 16:9, Portrait 4:5, Landscape 5:4, custom | Square 1:1 · 1024 |
| Count | 1 to 8 | 1 |
| Steps (Advanced) | 1 to 50 (8 to 50 for SDXL, Flux 1) | per model family |
| Guidance (CFG) (Advanced, SDXL) | 1 to 15 | 7 |
| Guidance (Advanced, Flux) | 1 to 10 | 3.5 for Flux 1 |
| Seed (Advanced) | number, or Random seed | random |
Model setup:
- Your models lists the folder, with Add a model (import a ComfyUI "Export (API)" workflow), Open folder and Change ComfyUI folder.
- A model missing its text encoder or VAE shows Required files with Download required files.
- An unrecognised model can be run with Try anyway.
- Changing to a different model family resets Steps and Guidance to that family's defaults.
Output: PNG named from the start of the prompt (a-red-fox.png, tag generated), saved to
your Downloads folder.
A grid of one-off PDF cards. Each card is one operation and accepts PDFs.
| Card | Does | Options (default) | Output |
|---|---|---|---|
| Extract text | Saves the text layer | none | name.txt (tag text) |
| Pages to PNG | Renders each page | Resolution, 72 to 400 dpi (150 dpi) | folder name (pages) with page-1.png, ... |
| Merge | Combines 2+ PDFs in table order | none; needs 2 or more PDFs | one PDF named after the first (tag merged) |
| Split | Keeps only the listed pages | Pages to keep, e.g. 1-3,5,8-10 (empty) |
name (pages).pdf |
| Burst | Saves every page separately | none | folder name (split) with name-1.pdf, ... (zero-padded to the page count's digits) |
| Extract images | Pulls out embedded images | none | folder name (images); fails if the PDF has none |
All cards use mutool, which is bundled.