A local catalog of Krea2-generated character and style samples, with searchable tags, reusable prompts, and an image registration and editing tool for Windows. English is the default interface language. Japanese is also available; your language choice is remembered.
This catalog is incomplete and provisional. Selection was based on visual comparisons, including color, shape, and overall impression. Some images were retained because they looked similar, rather than being exact reproductions. We could not thoroughly check every image by hand. Characters, names, tags, and styles may be wrong, and generated results may be inappropriate. Please recheck the downloaded collection and correct, hide, edit, or regenerate entries as needed. Corrections and suggestions are welcome.
Download Krea2-Character-Catalog-v0.9.zip for the complete application. It includes the accepted images, original tag CSVs, workflow examples, and Catalog Editor. Extract it to obtain a folder named Krea2-Character-Catalog. Python and generation models are not bundled.
GitHub's Code → Download ZIP and automatic Source code archives contain the development source, without the image collection or full tag dataset. Use the named ZIP attached to a Release for the complete catalog. Each release keeps its own versioned downloads. SHA256SUMS.txt provides checksums for the two ZIPs.
open_review.cmd.The preferred local address is http://127.0.0.1:8790/. If the port is occupied, the launcher chooses an available port and opens the actual address. It reuses an already-running service for this same catalog and does not replace other applications. The server listens only on the local PC.
The launcher checks for Python 3.10+, Pillow 10 to below 13, and WebP support. It prefers an existing compatible Python. When necessary, it creates a private .venv using a detected Python and installs dependencies there, without installing Pillow globally. The first dependency installation needs internet access. If Python is missing, install it from python.org.
To choose a specific Python, set KREA2_CATALOG_PYTHON, or run this from the extracted folder with your interpreter path:
powershell -NoProfile -ExecutionPolicy Bypass -File tools/open_review.ps1 -Python '<your-python-executable>'
Search filters, language, display settings, and page position are saved in your browser. Hidden entries, favorites, and star ratings are saved locally. The download starts without personal favorites, ratings, or editor drafts.
Select images to hide them together. Hide does not delete the image file. Use Hidden list, Show again, or Undo to restore visibility. Undo restores the entire previous hide operation. Export the visible catalog as CSV or the hidden list as TXT. Copyright sorting orders groups by the number of matches under the current filters.
See Catalog and Editor usage for the controls and saved-state behavior.
Krea2 Catalog Editor registers your generated images and lets you organize their name, copyright/category, reusable prompt, and tags. It does not generate images or automatically judge them.
open_editor.cmd.An image alone is enough. Every text field is optional. A blank name uses the filename, and a blank copyright uses Unclassified. The original file is copied without recompression.
Expand CSV / workflow JSON (optional) to search the supplied CSVs or your own CSV. For unfamiliar CSVs, choose the column mapping. Selecting a CSV does not register anything automatically. Workflow JSON is optional and can be removed. If the image already exists, choose the existing entry or explicitly add a separate entry.
Click a registered card to open its information panel, then click Edit to change its information inside that panel.
| Control | Result |
|---|---|
| Register / green Save changes | Saves the inputs to the catalog. |
| Blue Close (keep draft) / Esc | Closes while keeping a draft. Unsaved changes are not applied. |
| Orange Cancel (discard changes/inputs) | Discards unsaved inputs and the draft; saved information remains. |
| Clear text fields | Clears only the text fields and continues editing. |
Edits are saved separately and affect only changed fields. An empty registration file does not clear the existing catalog. Original images, generation records, visibility, favorites, and ratings are preserved.
| Location inside the extracted folder | Contents |
|---|---|
outputs/staging/<run>/images/ | Included accepted images, reusable prompts, and original tags. |
outputs/staging/<run>/generation_manifest.json | Links between included images and their records. |
outputs/user_items/<uuid>/ | Images and optional JSON added through the Editor. |
outputs/user_catalog.json | Added entries. |
outputs/user_catalog_edits.json | Changes to existing entries. |
outputs/review_state.json | Your hidden entries, favorites, and star ratings. |
original_tags/anima_characters.csv | Existing original character-tag CSV, included in the complete Release ZIP. |
original_tags/styles/selected_all_styles.csv | Optional style suggestions, included in the complete Release ZIP. |
comfyui/workflows/ | Shared ComfyUI workflow examples. |
lists/ | Accepted/rejected name and copyright lists, and review counts. |
The existing original-tag CSV uses tag,category,count,insert_mode. The tag field begins with character and copyright identities, followed by appearance tags. The generation input example, comfyui/inputs/thumbnail_prompts.csv, instead uses id,prompt. Reusable prompts, original tags, and the actual generation prompt serve different purposes.
For backup or moving to another PC, copy the entire catalog folder. Extract upgrades into a separate folder, keep the old version, and transfer your added images, edits, and review state. See Updating and backup.
See Generation and registration and ComfyUI workflows for the models, settings, examples, and required nodes. Set up your own generation environment, render an image, then register it through the Editor.
The detail panel offers Open original image and Show image in folder. Where a WebP contains generation metadata, drag it into a compatible ComfyUI installation to load the workflow.
In v0.9, the distribution copy removes the original PC's CSV paths from image metadata. Character workflows use the saved, resolved generation text through manual_prompt. Install or update the bundled ComfyUI-Krea2-Catalog-Tools under your ComfyUI custom_nodes folder and restart ComfyUI. Adjust model names and missing nodes for your environment. The compressed pixel data has not been changed. Some older images have no generation metadata, so not every image can reproduce its complete original settings.
Rejected images and blank rejected cards are excluded from the catalog ZIP. Their regeneration JSONs are available separately as Krea2-Character-Catalog-Rejected-Workflows-v0.9.zip. Of these, 13,486 are based on saved generation metadata, and 11 older entries are editable templates without original metadata. Running a rejected recipe does not guarantee an accepted result. You can use the recipes without adding rejected images back to the catalog.
The GitHub Pages site is a simplified read-only demo, not the full application. It displays prebuilt static data and does not run the catalog's Python backend. Its features and saving behavior differ from the local ZIP application.
Try searching, filtering, sorting, character/style switching, enlarging images, and copying prompts. Public data cannot be deleted, hidden, registered, or edited. Mutation controls and save APIs are absent. Use the local ZIP application to add your own images or change information. The online demo opens on Characters each time, even after a previous Styles session. Other demo display preferences are stored only in your own browser. The local application can remember its last tab.
The layout and organization of this catalog were inspired by AnimaDex. Normal registration and editing do not require AnimaDex or Codex.
For optional comparisons, install both catalogs locally and use the supplied krea2-animadex-review skill. Configure paths and ports for your own environment, reusing an existing AnimaDex installation. See Comparison setup and applying decisions.
AI suggests candidates; a human makes the final decision. Only selected, confirmed pass/fail decisions are applied. Unselected, held, or empty data does not alter existing catalog state. References are matched using CSV identity tags and both images, rather than a name alone. Identity matching by the skill is not itself AI visual judgment. The current comparison adapter supports characters; style references need separate handling.
| Type | Accepted | Rejected | Candidates checked |
|---|---|---|---|
| Characters | 3,809 | 13,122 | 16,931 |
| Styles | 110 | 375 | 485 |
| Total | 3,919 | 13,497 | 17,416 |
Counts refer to image candidates in the current reviewed set. Alternate outfits can be separate candidates. Retired or duplicate older style samples are excluded. There are no held or undecided candidates in this release scope. Accepted means provisionally selected, not guaranteed accurate reproduction. CSV lists contain names and copyrights/categories; internal record IDs and image hashes are omitted from these human-readable lists.
In these style checks, many artist styles associated with Pixiv or Danbooru were difficult to reproduce, while some well-known manga artists and illustrators had similar results. This is an observation from this catalog, not evidence establishing which artists were present in the model's training data.
Rights to characters, works, models, and third-party tag data remain with their respective owners. Check the applicable terms with each provider. To report a wrong name, tag, or generated result, include the entry name and catalog version in an Issue.
A local catalog of Krea2-generated character and style samples, with searchable tags, reusable prompts, and an image registration and editing tool for Windows. English is the default interface language. Japanese is also available; your language choice is remembered.
This catalog is incomplete and provisional. Selection was based on visual comparisons, including color, shape, and overall impression. Some images were retained because they looked similar, rather than being exact reproductions. We could not thoroughly check every image by hand. Characters, names, tags, and styles may be wrong, and generated results may be inappropriate. Please recheck the downloaded collection and correct, hide, edit, or regenerate entries as needed. Corrections and suggestions are welcome.
Download Krea2-Character-Catalog-v0.9.zip for the complete application. It includes the accepted images, original tag CSVs, workflow examples, and Catalog Editor. Extract it to obtain a folder named Krea2-Character-Catalog. Python and generation models are not bundled.
GitHub's Code → Download ZIP and automatic Source code archives contain the development source, without the image collection or full tag dataset. Use the named ZIP attached to a Release for the complete catalog. Each release keeps its own versioned downloads. SHA256SUMS.txt provides checksums for the two ZIPs.
open_review.cmd.The preferred local address is http://127.0.0.1:8790/. If the port is occupied, the launcher chooses an available port and opens the actual address. It reuses an already-running service for this same catalog and does not replace other applications. The server listens only on the local PC.
The launcher checks for Python 3.10+, Pillow 10 to below 13, and WebP support. It prefers an existing compatible Python. When necessary, it creates a private .venv using a detected Python and installs dependencies there, without installing Pillow globally. The first dependency installation needs internet access. If Python is missing, install it from python.org.
To choose a specific Python, set KREA2_CATALOG_PYTHON, or run this from the extracted folder with your interpreter path:
powershell -NoProfile -ExecutionPolicy Bypass -File tools/open_review.ps1 -Python '<your-python-executable>'
Search filters, language, display settings, and page position are saved in your browser. Hidden entries, favorites, and star ratings are saved locally. The download starts without personal favorites, ratings, or editor drafts.
Select images to hide them together. Hide does not delete the image file. Use Hidden list, Show again, or Undo to restore visibility. Undo restores the entire previous hide operation. Export the visible catalog as CSV or the hidden list as TXT. Copyright sorting orders groups by the number of matches under the current filters.
See Catalog and Editor usage for the controls and saved-state behavior.
Krea2 Catalog Editor registers your generated images and lets you organize their name, copyright/category, reusable prompt, and tags. It does not generate images or automatically judge them.
open_editor.cmd.An image alone is enough. Every text field is optional. A blank name uses the filename, and a blank copyright uses Unclassified. The original file is copied without recompression.
Expand CSV / workflow JSON (optional) to search the supplied CSVs or your own CSV. For unfamiliar CSVs, choose the column mapping. Selecting a CSV does not register anything automatically. Workflow JSON is optional and can be removed. If the image already exists, choose the existing entry or explicitly add a separate entry.
Click a registered card to open its information panel, then click Edit to change its information inside that panel.
| Control | Result |
|---|---|
| Register / green Save changes | Saves the inputs to the catalog. |
| Blue Close (keep draft) / Esc | Closes while keeping a draft. Unsaved changes are not applied. |
| Orange Cancel (discard changes/inputs) | Discards unsaved inputs and the draft; saved information remains. |
| Clear text fields | Clears only the text fields and continues editing. |
Edits are saved separately and affect only changed fields. An empty registration file does not clear the existing catalog. Original images, generation records, visibility, favorites, and ratings are preserved.
| Location inside the extracted folder | Contents |
|---|---|
outputs/staging/<run>/images/ | Included accepted images, reusable prompts, and original tags. |
outputs/staging/<run>/generation_manifest.json | Links between included images and their records. |
outputs/user_items/<uuid>/ | Images and optional JSON added through the Editor. |
outputs/user_catalog.json | Added entries. |
outputs/user_catalog_edits.json | Changes to existing entries. |
outputs/review_state.json | Your hidden entries, favorites, and star ratings. |
original_tags/anima_characters.csv | Existing original character-tag CSV, included in the complete Release ZIP. |
original_tags/styles/selected_all_styles.csv | Optional style suggestions, included in the complete Release ZIP. |
comfyui/workflows/ | Shared ComfyUI workflow examples. |
lists/ | Accepted/rejected name and copyright lists, and review counts. |
The existing original-tag CSV uses tag,category,count,insert_mode. The tag field begins with character and copyright identities, followed by appearance tags. The generation input example, comfyui/inputs/thumbnail_prompts.csv, instead uses id,prompt. Reusable prompts, original tags, and the actual generation prompt serve different purposes.
For backup or moving to another PC, copy the entire catalog folder. Extract upgrades into a separate folder, keep the old version, and transfer your added images, edits, and review state. See Updating and backup.
See Generation and registration and ComfyUI workflows for the models, settings, examples, and required nodes. Set up your own generation environment, render an image, then register it through the Editor.
The detail panel offers Open original image and Show image in folder. Where a WebP contains generation metadata, drag it into a compatible ComfyUI installation to load the workflow.
In v0.9, the distribution copy removes the original PC's CSV paths from image metadata. Character workflows use the saved, resolved generation text through manual_prompt. Install or update the bundled ComfyUI-Krea2-Catalog-Tools under your ComfyUI custom_nodes folder and restart ComfyUI. Adjust model names and missing nodes for your environment. The compressed pixel data has not been changed. Some older images have no generation metadata, so not every image can reproduce its complete original settings.
Rejected images and blank rejected cards are excluded from the catalog ZIP. Their regeneration JSONs are available separately as Krea2-Character-Catalog-Rejected-Workflows-v0.9.zip. Of these, 13,486 are based on saved generation metadata, and 11 older entries are editable templates without original metadata. Running a rejected recipe does not guarantee an accepted result. You can use the recipes without adding rejected images back to the catalog.
The GitHub Pages site is a simplified read-only demo, not the full application. It displays prebuilt static data and does not run the catalog's Python backend. Its features and saving behavior differ from the local ZIP application.
Try searching, filtering, sorting, character/style switching, enlarging images, and copying prompts. Public data cannot be deleted, hidden, registered, or edited. Mutation controls and save APIs are absent. Use the local ZIP application to add your own images or change information. The online demo opens on Characters each time, even after a previous Styles session. Other demo display preferences are stored only in your own browser. The local application can remember its last tab.
The layout and organization of this catalog were inspired by AnimaDex. Normal registration and editing do not require AnimaDex or Codex.
For optional comparisons, install both catalogs locally and use the supplied krea2-animadex-review skill. Configure paths and ports for your own environment, reusing an existing AnimaDex installation. See Comparison setup and applying decisions.
AI suggests candidates; a human makes the final decision. Only selected, confirmed pass/fail decisions are applied. Unselected, held, or empty data does not alter existing catalog state. References are matched using CSV identity tags and both images, rather than a name alone. Identity matching by the skill is not itself AI visual judgment. The current comparison adapter supports characters; style references need separate handling.
| Type | Accepted | Rejected | Candidates checked |
|---|---|---|---|
| Characters | 3,809 | 13,122 | 16,931 |
| Styles | 110 | 375 | 485 |
| Total | 3,919 | 13,497 | 17,416 |
Counts refer to image candidates in the current reviewed set. Alternate outfits can be separate candidates. Retired or duplicate older style samples are excluded. There are no held or undecided candidates in this release scope. Accepted means provisionally selected, not guaranteed accurate reproduction. CSV lists contain names and copyrights/categories; internal record IDs and image hashes are omitted from these human-readable lists.
In these style checks, many artist styles associated with Pixiv or Danbooru were difficult to reproduce, while some well-known manga artists and illustrators had similar results. This is an observation from this catalog, not evidence establishing which artists were present in the model's training data.
Rights to characters, works, models, and third-party tag data remain with their respective owners. Check the applicable terms with each provider. To report a wrong name, tag, or generated result, include the entry name and catalog version in an Issue.