Extension for Stable Diffusion WebUI Forge - Neo
Browse, download, and manage your CivitAI models directly inside Forge Neo — with auto-organization, disk usage dashboard, creator management, and support for all modern architectures (FLUX, Wan, Qwen, Pony, Illustrious, and more).
- What's New
- Changelog
- Roadmap
- Features
- Installation
- Auto-Organization System
- Dashboard & Statistics
- Supported Model Types
- Credits
- Full support for the new CivitAI domain split — CivitAI now separates SFW content (
civitai.com) from the complete catalog (civitai.red). The extension adapts automatically so nothing breaks. Paste any CivitAI link from either domain and it opens instantly. - New "SFW only" setting — a simple checkbox in Settings lets you restrict all links and API calls to
civitai.comif you prefer. Off by default, so the full catalog stays accessible without extra steps. - Update Mode filter isolation — Browser-tab filters are ignored when Update Mode is active, preventing "No updates match the current filters" false negatives.
pressRefresh()and page-slider triggers no longer pull Browser filters into the Update Mode view. - SHA256 silent-update detection — if a downloaded file fails hash verification, the extension re-queries the CivitAI API for the version's current SHA256. If the author updated the file silently, the new hash is accepted and metadata is updated instead of failing.
- Defensive ambiguity handling — searching by SHA256 that returns multiple candidates no longer crashes with
KeyError; the code safely falls back to an error message. - Exact search restricted to Model name — CivitAI API does not support quoted search for Tag or User name. Exact search now only wraps terms in quotes when "Model name" is selected.
- Batch download resilience — internal loop eliminates gaps between queued items, timeout increased to 30s with automatic retry (up to 3×), SHA256 buffer increased to 8MB.
- Update list sorted by mtime — outdated models are now listed with most recently modified first, making it easier to prioritize updates.
- Added centralized domain helper (
get_civitai_domain()) to replace all hardcodedcivitai.comURLs across the extension. - Added
civitai_sfw_onlycheckbox setting (default: off →civitai.red) to toggle between domains. - Fixed search-box direct-link parser to recognize both
civitai.comandcivitai.redURLs. - Updated all API calls, model page links, uploader profile links,
Refererheaders, and JSON sidecarmodelPageURLfields to use the configured domain. initial_model_page: verifygl.update_modebefore resetting state; ignore Browser-tab filters when Update Mode is active.download_create_thread: on SHA256 mismatch, re-query/api/v1/model-versions/{version_id}to detect silent file updates by the author.create_model_item/selected_to_queue: propagateversion_idthrough the download queue for post-download API recheck.resolve_ambiguity: defensively syncmodel_sha256andmodel_filenamewith the chosen candidate.- Hardened
initial_model_pageandupdate_model_infoagainst malformedgl.json_data(e.g. SHA256 ambiguity dicts withoutitems/metadata). - Exact search restricted to Model name only — CivitAI API does not support quoted search for Tag or User name.
- Batch download internal loop eliminates Gradio event-queue gaps between items.
- Timeout increased to 30s with automatic retry (up to 3×) for network failures.
- SHA256 verification buffer increased from 1MB to 8MB.
- Update list sorted by file modification time (most recent first).
- Fix re-trigger loop on
queue_triggerpreventing duplicate download executions.
- Browser version dropdown now defaults to an installed version whenever one exists, even when updates are available.
- This preserves delete action availability for models that are installed but outdated.
- Added a delete failsafe for card quick delete: if multiple installed versions exist, quick delete is canceled and the user is instructed to pick the exact installed version in the Browser panel before deleting.
- Installed-model Browser loading now keeps unmatched files visible as local-only entries while still using API data for matched models.
- Added automatic SHA256 cache sync for checkpoints right after successful download completion.
- Added a manual
Sync checkpoint SHA256 cachebutton in Update Models to reconcile local checkpoints against Forge cache. - Added local checkpoint hash registry (
lib/models/checkpoint_hashes.json) to track synced entries and clean stale records for deleted files.
- Fixed an issue where the local trigger word fallback process ignored API groups and flattened words into single lines.
- Fixed an issue where "Update model info & tags" didn't safely persist incoming
trainedWordsgroups natively to the local cache. - Added exponential backoff retry mechanism to API calls returning temporary 50x server errors, saving "Update models" loops from failing silently on affected files.
- Moved trigger word row buttons (📋 / ➕) to the left side of the text in the preview panel.
- Consolidated trigger words from
.safetensorsmetadata, local.jsonactivation text, and APItrainedWords - Added case-insensitive deduplication while preserving original order
- Model info now uses local consolidated trigger words first, with API fallback when local cache is unavailable
- Wan card badges now distinguish
I2V,T2V, andTI2Vsubtypes (API already returns specificbaseModelvalues) - New setting
civitai_neo_wan_subfolder_by_type(OFF by default): splits Wan downloads intoWan/I2V/,Wan/T2V/,Wan/TI2V/subfolders - Fixed multi-level subfolder "already organized" check (was using
os.path.basename, broke forWan/I2Vpaths) - Fixed Flux.2 Klein 4B/9B and Flux.2 D showing
F1badge — now correctly showsF2
- Each trigger word group gets its own row with individual copy and add-to-prompt buttons
- LORA tag row (
<lora:filename:1>) shown as first entry in purple/monospace - Clipboard copy with ✓ visual feedback (1.5s)
- "Add all to prompt" button when multiple groups exist
- Fixed wildcard base-model subfolder being applied in GUI even when
wildcard_organize_by_baseis OFF (civitai_api.pypath calculation was missing the wildcard guard) - Fixed delete-by-SHA256 silently failing —
json_basepath not joined withroot, makingos.path.existssearch in CWD instead of the model directory
- Own subfolder per wildcard download (sd-dynamic-prompts compatible)
- Flat zip extraction — no double-nesting when the zip has internal folders
- Skip preview/gallery images for Wildcards
- New settings:
wildcard_own_folder(ON by default),wildcard_organize_by_base(OFF by default)
- File integrity check — every download is verified against CivitAI's expected hash after completing; corrupted or incomplete files are detected and removed automatically
- Faster batch queueing — adding many models to the queue is now instant; metadata is fetched only when each download actually starts, not at queue time
- Safer cancel — cancelling a download or clearing the queue is now more reliable and won't cause hangs
- Aria2 auto-reconnect — if the download engine crashes during a session, it automatically restarts and resumes the current download
- Skipped model feedback — if a model can't be queued during batch selection, a visible error message appears in the Download Manager
- Download queue survives disconnects — the queue is saved server-side; if your session ends (e.g. RunPod timeout), a restore banner appears next time you open the UI
- One-click restore — re-queue all interrupted downloads with a single button click
- Companion files guidance — a warning banner appears when downloading models that require additional files (e.g. Wan 2.2, FLUX, Qwen), with specific instructions per architecture
- Background download fix — downloads no longer stall when the browser tab is in the background
- Favorite creators — mark creators as favorites; their cards get a gold glow and ⭐ badge
- Ban creators — hide cards from creators you don't want to see; one-click toggle
- Persistent lists — favorites and bans are saved to disk and restored across sessions
- Mutually exclusive — favoriting a creator removes them from the ban list and vice versa
- Export dashboard — download your collection stats as CSV or JSON
- Top 10 largest files — see which individual models take the most space
- Top 10 categories by file count
- Orphan detection — optional scan to find model files with missing or incomplete metadata
- Graceful 404 handling — when a model has been deleted from CivitAI, the extension shows the locally cached info instead of an error, with a clear warning banner
- LoRA activation syntax — "Add to prompt" now inserts the correct
<lora:filename:1>syntax automatically - SHA256 in model info — hash shown in the version details; click to select for easy copy
- Live progress bars — Validate and Fix operations now show real-time progress
- Validate organization — read-only scan that shows which models are correctly placed, misplaced, or missing metadata
- Fix misplaced files — move flagged models to the correct folders in one click; backup created automatically
- Instant undo — revert the fix immediately without scrolling to the Undo section
- "Add to prompt" button now works in the model info overlay (CivitAI icon in txt2img/img2img)
- Various crash fixes
- Shift+click on meta field buttons to append to prompt instead of replacing
- Deleted models go to the OS recycle bin by default
- Search settings now persist correctly across restarts
- Dashboard banner showing outdated model counts after scanning
- Batch update — select multiple outdated models and download all at once
- Version comparison by model family (not just version name)
- Retention policy on update: keep, move to trash, or replace
- Card color legend bar above the browser grid
- DNS retry on transient network failures
- Dashboard UX improvements (hide empty categories, scan summary)
- Full Forge Neo / Gradio 4 compatibility
- Auto-organization system with backup and rollback
- Extended architecture support
- Saved search presets
- Favorites in creator/user search
- Additional browser quality-of-life improvements
- Organization by Tag — Phase 1: save CivitAI tags to
.jsonsidecar; editable user-tags field in model panel for manual assignment - Organization by Tag — Phase 2: in Manage tab, pick "anchor" tags → models with that tag sort into
<type>/<tag>/subfolders (independent of base-model organization)
- All known issues resolved
- Full Forge Neo compatibility guarantee
⭐ = exclusive to Neo
- Browse CivitAI directly inside the WebUI — no tab switching
- Search by model name, tag, or username
- Filter by content type: Checkpoint, LORA, VAE, ControlNet, Upscaler, TextualInversion, Wildcards, Workflows, and more
- Filter by base model: SD 1.x, SDXL, Pony, Illustrious, FLUX, Wan, Qwen, NoobAI, Lumina, and more — list auto-updated from CivitAI at startup ⭐
- Sort by: Highest Rated, Most Downloaded, Newest, Most Liked, Most Discussed
- Filter by time period: Day, Week, Month, Year, All Time
- NSFW toggle, liked-only filter, hide installed models, hide banned creators
- Exact search mode
- Search settings persist across restarts ⭐
- Download any model, version, and file variant directly
- High-speed multi-connection downloads via Aria2 (optional, on by default)
- Download queue — multiple downloads run in sequence without blocking the UI
- Queue persistence — survives session disconnects with one-click restore ⭐
- Cancel individually or clear the entire queue
- Folder automatically set based on content type ⭐
- Custom sub-folders per download
- API key support for early access and private models
- Proxy support for restricted regions
- Orange border on cards with a newer version available
- Batch update — select multiple outdated models and download all at once
- Version comparison by model family (not just version name)
- Retention policy on update: keep, move to trash, or replace
- Dashboard shows outdated model counts after scanning
- New downloads automatically sorted into subfolders by base model (SDXL/, Pony/, FLUX/, etc.)
- Organize your existing collection in one click
- Validate organization — read-only check showing correct / misplaced / no-metadata per file ⭐
- Fix misplaced files in one click — automatic backup created first ⭐
- One-click rollback (keeps last 5 backups)
- Custom folder mapping in Settings
- Associated files (
.json,.png,.txt) always move with the model
- Model info panel with name, version, base model, type, tags, permissions, and description
- Sample images with "Send to txt2img" — fills prompt, negative, sampler, steps, CFG
- Individual meta field buttons — send just one field; Shift+click to append ⭐
- "➕ Add to prompt" in the model overlay — appends trigger words directly; auto-inserts LoRA syntax ⭐
- SHA256 hash shown in version info — click to select ⭐
- Video preview on hover for cards with video samples ⭐
- Save model info and images locally
- Disk usage by category and architecture
- Pie chart with percentage breakdown
- Top 10 largest files and categories
- Orphan file detection (optional)
- Export to CSV or JSON
- Update summary after scanning
- Color-coded borders: aquamarine = installed, orange = outdated, gold = early access / favorite creator
- Color legend bar always visible above the grid ⭐
- NSFW, Early Access (💎), and type badges
- Configurable tile size
- Quick delete from the card
- Multi-select checkboxes for batch download ⭐
- Favorite (⭐) and ban (🚫) creator directly from the card ⭐
- Deleted models go to the OS recycle bin by default (configurable)
- Filename sanitization — removes illegal characters automatically
- Filename length capped to prevent filesystem errors
- Open Forge Neo WebUI
- Go to Extensions → Install from URL
- Paste:
https://github.com/eduardoabreu81/sd-civitai-browser-neo - Click Install and reload the WebUI
⚠️ This extension requires Forge Neo. For Forge Classic or Automatic1111, use the anxety-solo fork.
The organization system sorts your models into subfolders based on their base model type, using the metadata saved alongside each file.
Before:
models/Lora/
├── model1.safetensors (SDXL)
├── model2.safetensors (Pony)
└── model3.safetensors (FLUX)
After:
models/Lora/
├── SDXL/model1.safetensors
├── Pony/model2.safetensors
└── FLUX/model3.safetensors
Enable "Auto-organize downloads" in Settings → Model Organization.
Go to Update Models tab → select types → click "📁 Organize models into subfolders".
- Automatic backup before any operation
- One-click undo
- Conflict detection (skips files that already exist at destination)
Go to the Dashboard tab, select the content types you want to analyze, and click "📊 Generate Dashboard".
You'll see:
- Total file count and disk usage
- Breakdown by category and architecture (Checkpoints and LORAs split by Pony, SDXL, FLUX, etc.)
- Visual progress bars and pie chart
- Top 10 largest files and categories
- Optional orphan file detection
Results can be exported as CSV or JSON.
| Architecture | Notes |
|---|---|
| SD 1.x / SD 2.x | Classic Stable Diffusion |
| SDXL | Base SDXL and derivatives |
| Pony | Pony V6 and variants |
| Illustrious | Illustrious XL |
| NoobAI | NoobAI (Illustrious-based) |
| FLUX | Dev, Krea, Kontext, Klein |
| Wan | Wan 2.2 — text/image to video |
| Qwen | Qwen-Image, Qwen-Image-Edit |
| Z-Image | Z-Image, Z-Image Turbo |
| Lumina | Lumina-Image 2.0 |
| Anima | Anima |
| Chroma | Chroma1-HD |
| Cascade | Stable Cascade |
| SVD | Stable Video Diffusion |
| Hunyuan | Hunyuan |
| Other | Catch-all; fully configurable |
Custom categories can be defined in Settings → Model Organization using a simple JSON pattern list.
- sd-civitai-browser by Vetchems — original project
- sd-civitai-browser-plus by BlafKing — foundation for this fork
- sd-civitai-browser-plus by anxety-solo — UI redesign and quality improvements
- sd-webui-civbrowser by SignalFlagZ — creator management inspiration
- Forge Neo by Haoming02
MIT — see LICENSE
Made with ❤️ for the Stable Diffusion community