feat: add sort-by-confidence option to WD14 captioning

[bmaltais]

Summary

Adds a Sort tags by confidence checkbox to the WD14 Captioning tab. When enabled, output tags are sorted by their model confidence score (highest first) before being written to the caption file, matching the ordering produced by the WaifuDiffusion online tagger.

• Default: unchecked (existing CSV-index ordering is preserved)
• Works independently of --character_tags_first and --always_first_tags/prefix

Backend change required in sd-scripts

This PR wires the GUI flag. The --sort_tags_by_confidence argument also needs to be added to sd-scripts/finetune/tag_images_by_wd14_tagger.py. The required change is:

In setup_parser() — add after --character_tag_expand:

parser.add_argument(
    "--sort_tags_by_confidence",
    action="store_true",
    help="sort tags by confidence score in descending order before outputting / タグを確信度の高い順に並べ替えてから出力する",
)

In run_batch() — replace the tag collection loop (lines ~255–272) with:

filtered_general: list[tuple[str, float]] = []
filtered_character: list[tuple[str, float]] = []
for i, p in enumerate(prob[4:]):
    if i < len(general_tags) and p >= args.general_threshold:
        tag_name = general_tags[i]
        if tag_name not in undesired_tags:
            filtered_general.append((tag_name, p))
    elif i >= len(general_tags) and p >= args.character_threshold:
        tag_name = character_tags[i - len(general_tags)]
        if tag_name not in undesired_tags:
            filtered_character.append((tag_name, p))

if args.sort_tags_by_confidence:
    filtered_general.sort(key=lambda x: x[1], reverse=True)
    filtered_character.sort(key=lambda x: x[1], reverse=True)

for tag_name, _ in filtered_general:
    tag_freq[tag_name] = tag_freq.get(tag_name, 0) + 1
    general_tag_text += caption_separator + tag_name
    combined_tags.append(tag_name)

for tag_name, _ in filtered_character:
    tag_freq[tag_name] = tag_freq.get(tag_name, 0) + 1
    character_tag_text += caption_separator + tag_name
    if args.character_tags_first:
        combined_tags.insert(0, tag_name)
    else:
        combined_tags.append(tag_name)

Notes

• No agent brief was written for this issue — implementation derived from triage analysis
• The GUI's always_first_tags field uses add_pre_postfix (not --always_first_tags); this pre-existing behavior is unchanged

Closes #3491

[github-advanced-security]

CodeQL found more than 20 potential problems in the proposed changes. Check the Files changed tab for more details.

[Copilot]

Pull request overview

This PR adds a WD14 Captioning GUI option to sort output tags by model confidence (to match WaifuDiffusion’s online tagger ordering), and also includes several additional changes across captioning utilities, server launch configuration, TensorBoard detection, docs, and dependency/config updates.

Changes:

• Add a “Sort tags by confidence” checkbox to the WD14 captioning tab and forward it as a CLI flag.
• Add Gradio server allowed_paths support via config.toml, and document it.
• Refactor manual captioning UI and add a new “Kontext Manual Captioning” tab, plus related dependency/tests/scripts updates.

Reviewed changes

Copilot reviewed 19 out of 21 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
uv.lock	Adds py-cpuinfo to the locked dependency set.
requirements.txt	Adds py-cpuinfo dependency.
requirements_macos_arm64.txt	Adjusts pip options / removes an entry to align macOS ARM requirements.
requirements_linux_ipex.txt	Updates IPEX/XPU package pins and indexes.
README.md	Updates docs links/TOC and documents allowed_paths under “Server Options”.
pyproject.toml	Adds py-cpuinfo dependency.
local_train.py	Adds a local training/setup helper script.
kohya_ss_colab.ipynb	Adds a Colab notebook for launching via gui-uv.sh.
kohya_gui/wd14_caption_gui.py	Adds the “sort by confidence” checkbox and forwards --sort_tags_by_confidence.
kohya_gui/utilities.py	Registers the new “Kontext Manual Captioning” tab in Utilities → Captioning.
kohya_gui/manual_caption_gui.py	Refactors manual captioning UI/state handling and import logic.
kohya_gui/kontext_manual_caption_gui.py	Adds a new manual captioning workflow for paired control/target images.
kohya_gui/class_tensorboard.py	Changes TensorBoard visibility detection to depend on tensorboard binary + AVX CPU support.
kohya_gui/class_gui_config.py	Adds allowed_paths config extraction and stores it on the config object.
kohya_gui.py	Passes allowed_paths into gradio.Blocks.launch().
gui-uv.sh	Adds commented guidance for optional bitsandbytes source builds.
config example.toml	Documents new [server].allowed_paths configuration key.
tests/test_tensorboard_visibility.py	Adds unit tests for TensorBoard visibility/AVX detection logic.
test/test_allowed_paths.py	Adds an integration-style test attempting to validate allowed_paths.
.github/workflows/typos.yaml	Bumps crate-ci/typos GitHub Action version.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bmaltais]

Closing — the GUI change alone is insufficient since the --sort_tags_by_confidence flag doesn't exist in sd-scripts yet. The required backend and GUI code is documented here for reference if this is picked up in sd-scripts in the future.