Skip to content

Features

Full capability list for Castcut, grouped by area. For a short overview, see the main README.

Jump to: Prompt generation · Scene tools · Studio · Gallery · Queue & ComfyUI · Workflows · Settings & data · Auth & API · Automation · UI & UX

Prompt generation & models

  • Searchable model picker with architecture-family filters
  • Profile-based prompt styles shared across related checkpoints
  • Prompt detail levels (Concise / Balanced / Rich) with combined model × detail size limits
  • Minimum character enforcement for long-form models (Image-2.0 Rich, FLUX Rich, etc.)
  • Positive and negative/preserve prompt modes
  • Uncensored system prompts (no content filtering or refusals)
  • One-click copy for ComfyUI paste
  • LLM-powered generation/formatting with rules fallback
  • Prompt diagnostics — lint sport/duo/helmet conflicts before or after generation
  • Compact & cross-model reformat — trim to model limits or reformat for the alternate model from any result panel
  • Pre-lint + rule fix — Duo shows hint lint before generate; Fix prompt (rules) applies helmets, sport strips, etc.
  • Qwen Edit builder — segment-based edit instruction builder on Generate for Qwen Edit models
  • Generate sport presets — sport preset chips on Generate (positive mode) with shareable scene URLs
  • Model recommender — Generate sidebar suggests models from prompt text (/api/models/recommend; UI uses server route with client fallback)
  • Token / weight inspector — (tag:1.2) analysis on Lint, Format, and result panels
  • Prompt readiness score — pre-queue lint badge on result panels (/api/readiness)
  • Readiness-gated queue — result panels warn below score 60; confirm or fix before ComfyUI queue
  • Readiness auto-fix — one-click compact / rule-fix / reformat from readiness panel
  • Negative prompt learner — learns tokens from low gallery ratings; Settings → Advanced
  • Avoidance preview — Settings shows matched tokens and LLM instruction before generation
  • Avoided tokens — low gallery ratings record motifs to avoid; all generators honor avoidance via LLM instruction and template pool filtering
  • Avoided tokens settings — view, add, remove, and clear the local avoidance list in Settings
  • Avoided tokens import/export — JSON list management on Settings
  • Context-aware negatives — ComfyUI queue picks negative profiles from tool/hints (pet, fantasy, sport, etc.)
  • Auto-negative on queue — optional Settings toggle + negative profile library for SD-family ComfyUI queue
  • Lint playground — /lint for paste-and-fix without generating a new scene

Scene tools & catalogs

  • Mobile Studio — /m (Add to Home Screen via manifest-mobile.json): phone-first film loop — Cast → Look → Outfit → Day → Story (Queue / Gallery under More). Touch shells on /m/moodboard, /m/fitting, /m/day, and /m/story reuse desk orchestration (swipe kits, four Day slots, Story stills and clips via I2V/T2V/continue). Cut film / Save to Cast are primary on phone (shared assembler, server ffmpeg preferred). Header Desk bridge is an optional large-screen handoff to Film (/play).
  • Play workspace (first-run default, Make) — sidebar footer / Profile → Appearance: Film, Look, Outfit, Day, Story, Gallery, and Queue. Studio is Control; Full is Build. Welcome asks What do you want to make? (Character / Scene·Film / Image / Surprise) before Heal & ready. Look lives under Scene nav and ⌘K (/moodboard). Simple remains available for a lean essentials layout.
  • Simple Essentials — Dashboard, Generate, Cast, Story, Gallery, and Queue; Studio, Video, Refine, Compose, Inpaint, Outfit, Day, Look, and the rest sit under More tools (⌘K finds everything).
  • Gallery collections — bulk Collect (favorite/rate), Group (named batches), and Project (campaign scope); tags are under Project in full mode only.
  • Lean gallery — Simple and Story workspace modes use slimmer filters and bulk actions on /gallery.
  • Outfit — /fitting (Play workspace / Scene nav): lock a Cast character plate, then one Clothing row (kit or your photo · shoes) whose Choose… opens the Clothing sheet shared with Day and Story: the kit deck filtered by clothing type (quick-pick chips — outfits, tops, bottoms, outerwear, formal, swimwear — plus More types…) with a large Selected kit preview and Prev / Next, BYO clothing (extract a clothing photo to a packshot, upload a ready packshot, or Save for later — durable browser KV, shared with Day/Story), Footwear (kits with packshots, your own shoes — worn photo or packshot, with Rescan — or your own words, named on every clothed still and shown beside the clothing on Qwen Edit 2511) and Draft previews (Boogu Edit Turbo thumbs when available). Quality is one preset — Fast / Balanced / Best (Balanced = Good render + Front and back; Best adds Auto-review try-ons: face match vs plate + vision outfit score, Best match marker, scores only) — with the switches and Good / Best by hand under Advanced. Pose — as the plate, a Custom pose from the pose editor (drag joints, turn in 3D, bend at the waist, start from any Day pose or a real reference pose, My poses, a partner figure), or the pose From a photo; the try-on prompt opens with the pose in words. Each try-on is followed by a back view beside it, and with shoes picked the feet are checked and a shoe pass puts them on when they came out wrong. Queue shows a clear block reason without Cast/plate/kit/BYO/isolate. Plate → Try-on → Keep → Day phase strip + status line (plate · kit/BYO). Keep on each compare card and in the lightbox; Pass and Requeue · new seed in the card's ⋯ menu (Pass dismisses a try-on; Skip kit advances the deck). Keep saves the outfit to the Cast's active look (its tile shows it; switching looks restores it), hands it to Day as the Day's outfit, and surfaces a primary Continue to Day CTA; Skip outfit · Day on desk + phone. Empty clothing-type filters offer Show all types. Deep links ?character= and ?wardrobe=; Save kit to Cast writes lockedWardrobeId; Continue in Story unlocks after the first Day film. Cross-links to Day and Look. Look packs apply via ?from=look. Mobile can upload / Gallery-pick a plate in-place.
  • Day — /day (Play bottom bar / Scene nav): pick a Cast character and plan four time slots (Morning → Night) with wardrobe kit, setting, and beat per slot. The page is a board: a one-line plan bar (Stills · Mood · People · Weather · Quality · Advanced), the slot cards, one action row, the reel; everything about one slot opens in its slot sheet (right drawer on desk, bottom sheet on a phone) from the card's Edit or its one ⋯ menu, and Setup (Cast, look, plate) is a status chip beside the title. Suggest day fills Setting/Beat (Queue keeps your plan); per-slot Reroll plan; Queue shows a clear block reason without Cast/plate/isolate. Sticky Ready to cut can be Hide’d (Cut film stays on Day reel). Quality is one preset — Fast / Balanced (default: Best render + Face finish + Redo pose misses once + Pick the best engine per pose) / Best (adds Auto-review, Best of two, Face boost) — with the seven switches, Good / Best by hand, the Pose pack and Day notes under Advanced ▾. A Look & clothing row under the plan bar sets the Day's look, kit / photo and shoes for every slot (a slot's sheet can still pick its own). BYO clothing + Image 3 pose stick-figure for stance unlock (Image 3 Edit only — pose guides do not attach ControlNet; everyday stills default to one adult; People in the plan bar opens Mixed / Duo companions). Mood (plan bar): Everyday / Suggestive / Sport / Vacation / Intimate / Raunchy (Intimate + Raunchy require NSFW env; Mixed / Solo / Duo under both adult moods). Intimate/Raunchy Duo prefers missionary / bent / spoon / wall beats (less lap-straddle merge), stronger nude discard, and HANDS / skin-texture locks. Suggestive uses heat-only beats (lingerie/unzip/charged poses) that own the stance — everyday walk/coffee baselines no longer flatten the mood. Sport picks a random sport that fits that time of day with a mid-action pose and athletic kit lock. Raunchy Solo uses wardrobe-fail setup + crude self-touch punchline (not tame flash pin-ups). Slot cards show Setting · Beat (or empty hints); pose/camera/beat pools rotate for variety. Phase + status strip (plate · mood · stills/clips + queue block). Look→Day / Outfit Keep→Day auto-align adult Setting/Beat (anti-book). Each Cast keeps its own Day — switching Cast sets the Day aside and brings it back when you return (the last six). Queue one slot (from its sheet) or the whole Day. After stills, Animate → Cut coach; Animate slot / Animate all queues I2V clips from completed stills. Completed stills and clips sync into a Day reel (clips preferred); Cut film prefers server ffmpeg H.264/AAC MP4 via /api/film/assemble (widescreen / vertical / square, 720p/1080p, optional crossfade/audio bed, clip trims) and falls back to browser MediaRecorder; before encoding /api/film/check reads each shot's file (missing, black frames, a talking clip with no sound) and the pre-cut list adds them to face / pose / Auto-review misses; each finished still and each Cut shot carries a face-match mark (Looks like Nora / Face may drift / Not Nora?); auto-stamps when Cast is set, and surfaces first-cut Watch on Cast / Open in Gallery (derivedKind=film). Deep links ?character= and ?wardrobe= (seeds empty slot kits); Look packs via ?from=look. Cross-links to Outfit, Look, and Story (?from=look when a pack is staged).
  • Day / Story clips on LTX-2.5 — with LTX-2.5 picked, every Animate clip runs on it except two-person adult clips (WAN unless Use LTX-2.5 for two-person adult clips too); one-person and adult clips hold the still harder (0.85) at 1024 for a closer face; the Animate sheet shows the default direction and formats your own for the engine; opt-in Refine detail (Lightricks Refine-Details IC-LoRA pass). Talking clips keep a wide still wide and her pose as it is, keep the scene moving (steam, water, leaves, crowds named by the place and line) and mix the place's room sound under the voice (a blurred-picture sound pass, level-matched by castcut-nodes 1.7.0); eleven ways to say a line.
  • Look — /moodboard (Play nav / Scene / ⌘K): stack up to four reference tiles (mood, lighting, location, style, palette) with optional gallery stills; choose a merge template. Tiles → Extract → Plate → Continue phase strip + status line (tiles · plate · pack). Extract/Queue show a clear block reason. Preview prompt is text only; Queue scene is optional (single primary Queue — result panel Queue hidden). Extract look builds a session look pack and soft-advances to Outfit with Go to Day instead. Skip look · Day (desk + phone). Upload / Gallery look plate in place; after a scene still, Keep as plate or requeue. Continue to Outfit / Day / Story hand off vibe notes + optional wardrobe lock. Save on Cast / Export JSON / Share this look under More. Mobile empty boards offer Add starter tiles. Deep link ?character=; cross-links to Day, Outfit, Story, and Film (/play).
  • Film (Play hub) — /play · /m/film: Cast → Look → Outfit → Day → Story (optional after first cut). Empty hub shows a sample reel + starter; Cast create is name + From photo first (traits under More). Mid-film has one Continue (Restart secondary); complete leads with Watch / Same look new Day / Story. Phone remaps every handoff to /m/...; Identity ready + compact persistence on the hub. Export / Import JSON / share link under Look packs; durable resume via play-campaign-v1.
  • Story — /story (Play workspace, or More tools in Simple; unlocked after first Day film): continues an existing Cast lead (Part / From photo / look plate live on Film + Cast — Story does not re-cast). Beat picker sits above the reel with pinned Setting/Tone/Content, adult Solo/Duo/Mixed, queue block reasons, Animate→Cut coach, and lightbox requeue; each reel card has one ⋯ menu (Edit scene, Pose…, Open full size, Queue / Animate / Play another, Copy prompt, Open in ComfyUI) and a scene sheet for editing its text and pose. Phase + status strip (Cast · plate · beats/clips). Write a bio, roll four story beats that continue from the last pick (unpicked cards stay off later rolls), tap a beat to queue a still or a clip in the story reel. Accepts Look packs via ?from=look&character= (seeds Setting, extra hints, inferred Tone, and wardrobe lock). Outfit for stills (desk + phone): catalog kit or BYO packshot as Image 2; photo stills attach an Image 3 pose wireframe from beat text (intimate + social layouts), with photoreal / anti-merge locks and intimate euphemism→anatomy rewrite before queue; pose is Image 3 Edit only (no ControlNet on filled mannequin guides). A still becomes I2V; a text-only beat is T2V; later beats continue from the last clip — Extend clip (Fal LTX extend-video, Grok /v1/videos/extensions, or Runway /v1/video_to_video), Continue from last frame (Replicate last-frame I2V; Fal soft-fails CDN upload → last-frame), or Stitch continue (Gemini last-frame I2V then server /api/film/assemble concat for the watch playlist). Story/Video surface which path ran. Cut film prefers server ffmpeg via /api/film/assemble and falls back to browser MediaRecorder; Save to Cast reuses that assembled blob (no second encode) when the cut succeeded. After cut: first-cut Watch on Cast / Gallery deep-links (manual — no auto soft-advance). Identity for this story From bio (default) or From photo (Cast look plate / upload). Optional Setting seeds a location into bios, beats, and stills. Exclusive Tone and Content rating plus an additive Gore toggle. Stills and clips render inline; download a zip of story.md + stills. Story sticks to the active Cast member (no separate session shelf). Consenting adults only; no minors. Default PG-13.
  • Cast — /characters: character homes for looks, stills, clips, assembled Films, Part, look plate, bible, and the Story film cut. Status strip (plate · Part · looks · films) warns when Outfit/Day need a plate; Start a film stays primary with Look/Outfit/Day/Story beside it. A What's next checklist tops the page (plate, traits, bible, outfit, Day, film, Story). The home is split into tabs — Overview (the Looks strip: one picture tile per look, New look, Prepare plate), Bible (Appearance traits, Story rating, Part + bible, Describe from photo, Picture this bible and Gender swap — the rendered picture as the Cast's other-gender version, same scene and outfit, shown beside it; pasting a whole bible is folded away), Film & media (Film studio with encode settings under Cut options, plus Media), and Packs & LoRA; ?media= opens Film & media and ?tab= any tab. Plate check — the look plate is read with DWPose (comfyui_controlnet_aux) plus its pixel size and flagged when it has no person or face, a face turned away, more than one person, a face under ~80px wide, or a short side under 512px; off with a note when the node pack is missing. Media tiles whose clip can't load (ComfyUI offline, file gone) show Clip unavailable instead of a broken image. Uploading a look plate soft-advances to Outfit (cancel or Go to Day). Roster shows No plate readiness. Continue reel uses Video I2V; Continue in Story works for Play/Outfit Cast records (synthesizes Story state from Cast bio/descriptor when needed). Roster empty CTA points at Create on Film. Look plate Choose from Gallery / upload / Remove write body plate + face lock together (Remove clears every look). Phone Watch (?media=films) keeps Cast name + Day/Story/Outfit jumps. LoRA flywheel — Export → Train writes keepers under PROMPT_DATA_DIR and starts /api/lora-train with datasetPath; Register & pin installs into Comfy models/loras; Prove it (or auto on register) queues a validation still.
  • Compose — /compose: multi-image transfer / modify with identity lock, regional edit, and gallery re-edit. Isolate on white (default) cuts Image 1 onto a blank plate so the original scene does not leak; Images 2–4 stay intact as pose/scene donors. Continue-edit gallery handoffs skip isolation so the full canvas remains. Transfer scan vision-reads every filled slot and writes a transfer instruction (recipes: pose↔people, people→pose+scene, outfit, background, style/light, hair, expression). Z-Image sends Figure 1 to the sampler; extras stay in the prompt. Cloud img2img is Image 1 only unless the model is a documented multi-ref edit (Fal Kontext multi / FLUX.2 edit / nano-banana edit, or Replicate multi-image Kontext). Transfer stays blocked on single-ref cloud. Identity lock: local Comfy uses IP-Adapter / InstantID / PuLID; Diffusers stills compile SDXL IPAdapterAdvanced natively (FaceID/PuLID drop-ins fall back to hub Plus weights; InstantID compiles natively on SDXL for txt2img/img2img/inpaint; PuLID stays Comfy); cloud uses face-ref (session face + Image 1 as ordered multi-refs when the endpoint allows, otherwise a weighted identity prompt — never Comfy IP-Adapter on cloud).
  • Character generator — solo, duo/sport, and compose-with-background modes; sport presets, team kit, batch roll, ComfyUI queue
  • Scene compose mode — Character tool merges background + subject into one scene prompt
  • Regional prompt composer — Character tool merges labeled subject/background/lighting segments; queue-time workflow-regional-patch also wires AttentionCouple-style nodes when installed
  • Location blocklist — block locations in Studio catalog; all generators respect the list
  • Locked wardrobe — pin a catalog outfit from Studio; Character and batch tools reuse it
  • Locked location & variation seed — pin scene place and environment seed for reproducible rolls
  • Scene starter catalog — ~294 searchable presets on Generate/Character (category, framing, tag filters; / focuses search)
  • User scene starter presets — save current hints or promote gallery analytics tokens; export/import starter packs in Studio Presets
  • Hint source — Manual, From history, or Random on Generate, Character, Pet, Fantasy, Background, Topics, and Variations
  • Persisted preset filters — search, framing, and tag filters survive reload via settings cache
  • Use as hints — Studio history rows open the source tool with hints prefilled (hintSource=manual)
  • Queue from preset — selected scene preset → Queue 4 variations handoff to /variations?from=preset
  • Active character descriptor — shared mandatory character sheet injected into Character API requests
  • Wardrobe avoided tokens — low-rated motifs filter catalog wardrobe picks across generators
  • Catalog rating bias — Studio catalog sorts clothing/locations by gallery review scores; click Insert to add to hints
  • Rating-driven random — history/gallery favorites and downvotes subtly adjust random-scene wildness
  • Adult generator plugin — env-gated /plugins/nsfw-generator with 120+ built-in presets, search/favorites/recent, user-saved presets, and Rapid AIO / Z-Image / Qwen quality recipes. A one-time in-browser acknowledgment (18+, legal where you are, consenting adults only) gates the tool itself, separate from the env flags that gate the deploy.
  • LoRA stack inline tuning — per-run model/clip strength overrides in the sidebar (session-only; included in session recipes and compare recipes)

Studio

  • Studio — prompt history with ratings, model compare, catalog browser, templates
  • Studio presets & diff — named scene lock bundles, history search/filters/tags, word-level prompt diff, custom templates, shareable ?scene= preset URLs
  • Studio analytics — per-user history + gallery activity summary and gallery rating token stats (high vs low motifs)
  • Studio analytics actions — add negative-scoring tokens to avoidance from the Analytics tab
  • Analytics live refresh — Studio analytics updates when gallery ratings change
  • Studio deep links — /studio?history=<id> highlights a saved prompt; gallery links back to history
  • Prompt projects — named campaigns filter Studio history; new saves attach active project id
  • Prompt iteration tree — Studio tab shows parent/child history branches via parentHistoryId
  • Iteration tree actions — Regenerate, Refine, and re-queue from iteration tree nodes
  • Iteration tree export — download parent/child history branches as structured JSON
  • Iteration branch diff — compare parent/child prompts on Studio iteration tab
  • Auto lineage — Improve/Refine/Reformat saves attach parentHistoryId for Studio iteration tree
  • Refine diff panel — word-level diff when refining from a saved history parent
  • Cherry-pick merge — Studio Diff tab merges two prompts with lint checks
  • Experiment dashboard — Studio Experiments tab groups gallery outputs by prompt/seed variants
  • Experiment list virtualization — window virtualizer for large experiment group lists (48+ groups)
  • Shared-project collab — presence bar + field-level draft sync (hints, instruction, positive, etc.) via BroadcastChannel and /api/collab SSE; room state persists to SQLite and optional COLLAB_REDIS_URL for multi-node; Settings → Overview shows collab backend health; Apply draft merges remote edits on Generate, Character, Refine, and Compose
  • Experiment winner workflow — crown winners, compare export, re-queue groups on Studio Experiments tab
  • Style transplant — Studio → Experiments applies lighting/camera mood from one prompt to another
  • Duplicate detection — Studio → Experiments finds near-identical history clusters
  • Multi-model portfolio — Studio Portfolio tab formats one draft for several models and batch-queues
  • Portfolio diff export — cross-model Markdown/HTML diff from Studio Portfolio tab
  • Portfolio CLI queue — npm run prompt:cli -- portfolio --input "..." --queue formats and queues each model to ComfyUI
  • Visual model compare — Studio Compare tab can queue both models to ComfyUI and show outputs side-by-side
  • Character identity bundles — export/import reusable character sheets from Studio Presets
  • Character identity bundles with saved list — browser-local list (descriptor, LoRA triggers, IP-Adapter ref) alongside JSON export/import
  • Preset packs — import/export bundles of scene presets from Studio Presets tab
  • Prompt brief — export/import portable prompt bundles from Studio Presets
  • Campaign templates — save/load campaign recipes on Studio Campaign tab
  • Studio campaign runner — batch random scenes or topics with optional ComfyUI queue
  • Project bundles — export/import project history + gallery JSON from Studio Projects tab
  • History/gallery export — CSV and JSONL export from Studio and Gallery bulk actions
  • History batch re-queue — re-queue saved batchPrompts from batch ComfyUI sends
  • History bulk actions — tag or delete all entries matching the current Studio history filters
  • Semantic search — token-overlap ranking in Studio history and Gallery filters
  • Embedding search — semantic history filter uses Ollama embeddings when available (/api/search/embeddings)
  • Global search — command palette (Ctrl+K / Ctrl+Shift+K) searches history, gallery, and scene presets
  • Gallery — /gallery stores queued jobs in IndexedDB (Dexie) and displays output images, clips, audio, and 3D meshes when ComfyUI finishes; previews appear inline on result panels
  • In-place clip playback — WAN/LTX animated WebP and mp4/webm play on gallery cards, lightbox, Story beats, Video results, Queue, and Cast instead of a flattened still; view proxies skip Sharp flattening for motion bytes
  • Audio and 3D gallery media — Stable Audio wav/flac/mp3 and Hunyuan3D glb/gltf/obj are captured from Comfy history, persisted as originals, playable (audio) or downloadable (3D) from cards/lightbox, and included in ZIP export. Filter chips: Audio / 3D.
  • Import completed host jobs — Gallery import prefers Comfy GET /api/jobs and falls back to /history; walks every pool host; imported rows keep the workflow graph when present
  • Failed-job node install — missing class_type errors offer Install missing nodes (Manager + restart + retry) on Queue, Gallery, and workflow preview
  • Gallery stats bar — at-a-glance totals (completed, in queue, favorites, unreviewed, avg rating) with one-click filter chips
  • Gallery layout modes — Grid, Dense, or List view (persisted)
  • Gallery review focus — review mode auto-selects the first card; keyboard 1–5 / F / N / P
  • Gallery review mode — rate completed outputs 1–5; low ratings feed wildness/avoidance bias
  • Gallery review shortcuts — 1–5 rate, F favorite, N/P navigate in review mode
  • Gallery review auto-advance — optional jump to next unreviewed item after rating
  • Mobile gallery review — touch-friendly rating bar in gallery review mode
  • Gallery compare modal — compare 2–4 selected outputs in a full-screen overlay (GalleryCompareModal + useGalleryCompareHandlers)
  • Exact graph replay — queue stores capped workflowJson on gallery entries; cards show an Exact graph badge and Replay exact graph (falls back to Comfy history when omitted)
  • Lineage filters — Upscale / Refine / Soft pass / Variation / Moiré / Face detail / ControlNet derivative chips
  • Gallery compare panel — pick winner, rate, favorite, mutate, or improve; bulk Seed experiment
  • Gallery compare — select 2–4 completed entries for side-by-side review on /gallery
  • Gallery card polish — hover quick actions (Open, Improve); storage cap warning near 5,000 IndexedDB entries
  • Gallery tools — favorites, status/tool/media filters, image/audio/3D download, sidecar JSON export per entry
  • Gallery custom groups — select cards → Group menu to assign a name; Groups rail to browse; Rename / Delete group clears the label across items. Filter by Group (or Ungrouped); click a card badge to show that set.
  • Gallery project filter — filter /gallery by the active Studio project
  • Gallery project filter & assign — filter by project dropdown; bulk assign entries to projects
  • Gallery ZIP export — bulk export selected entries as originals (images, clips, audio, meshes) + sidecars
  • Gallery → Refine / Image→Prompt — open completed outputs with image + prompt pre-loaded
  • Gallery handoffs — send selected prompts to Topics batch or Variations matrix (?matrix=1)
  • Gallery param grid — CFG × steps experiment grid from a selected entry
  • Gallery A/B export — export compare selections as JSON or HTML side-by-side reports
  • Gallery embedding search — semantic and find-similar use /api/search/embeddings when available
  • Find similar outputs — rank gallery entries by prompt similarity to a selection
  • Mutate winner — re-queue gallery entries with location/wardrobe/wildness/variation mutations
  • Compare winner lineage — pick winner sets lineage parent when the entry has history
  • Compare pick-winner auto-improve — high-rated winner triggers auto-improve loop
  • Param experiment queue — sweep CFG, steps, width, or seed from a selected gallery output
  • Vision gallery review — AI-suggested rating, tags, and critique in review mode
  • Gallery vision tags — auto-tag completed outputs; filter by “Vision tags”
  • Fullscreen slideshow — gallery filter bar starts immersive slideshow with keyboard controls
  • Improve output pipeline — one-click Improve from gallery or result panels opens Refine with image + intent
  • Low-rating refine loop — 1–2★ gallery ratings open Refine with corrective intent automatically
  • Intimate Solo heat — solo fingering beats use wild stances; Image 3 wrists at the pelvis; Raunchy Solo is fingers-only mid-self-touch that follows the beat stance. Nude Day stays on Edit NSFW; edit lead uses natural language: indoor SETTING (no beach), clothes-are-now-gone, then the beat/Image 3 pose named explicitly; Rapid 4–8 steps CFG 1 euler_a/simple; face crop ≥1.1MP when Cast face duplicates lingerie.
  • Gallery skin refine — Soft-pass from Gallery card / compare (default Klein 9B Base ReferenceLatent edit, denoise 1); keeps navel/moles while killing oily plastic. Model picker in Settings → Auto-improve. Not auto-queued after Day/Story stills.
  • ELO tournament — bracket compare mode in gallery compare
  • Aesthetic scoring — heuristic gallery score on cards; POST /api/aesthetic/score for snapshots
  • Gallery PWA — optional service worker caches /api/comfyui/view image responses for faster revisits
  • PWA manifest — installable web app metadata (manifest.json)

Queue & ComfyUI

  • Workflow takeover — at queue time auto-bind placeholders, patch latents/loaders/samplers, insert FLUX sampling nodes, and upscale outputs; see Workflow takeover
  • Queue quality profiles — sidebar Draft / Final / Max; per-tool overrides in Settings; gallery stores and re-queues with stored profile
  • ComfyUI job status — polls ComfyUI history after queue; pending/running/completed in the UI
  • ComfyUI WebSocket progress — optional faster job updates in Settings
  • Re-queue — gallery entries and Studio history can be sent to ComfyUI again (same params, new seed, upscale, or new variation at Final/Max)
  • Variation grid — /variations rolls N prompt variations and batch-queues them with unique ComfyUI seeds
  • Prompt matrix mode — /variations?matrix=1 for row×column variation grids
  • Matrix CSV export — Variations matrix mode exports row×column grid to CSV
  • Batch ComfyUI queue — queue all duo batch rolls to ComfyUI with shared negative
  • Topics batch build — turn a topic list into full Generate or Character prompts via /api/topics/batch
  • Topics → Variations handoff — send a topics batch to /variations as an imported grid
  • Batch lint gate — Topics and Variations bulk queue lint prompts first (fix-all or skip errors)
  • Batch history auto-save — batch ComfyUI queue saves one lineage history entry when auto-save is enabled
  • Export pipeline — “Prepare for ComfyUI” runs lint → fix → compact → copy pair → optional queue from any result panel
  • Preview workflow — dry-run before queue on Generate, Character, Format, Lint, Refine, and other result panels
  • Workflow dry-run — preview injected workflow JSON in Settings (and from Lint result panels) before queueing
  • Custom workflow tokens — user-defined placeholders like {{CHECKPOINT}} and {{LORA}} with values in Settings
  • ComfyUI workflow — optional COMFYUI_WORKFLOW_PATH or Settings → ComfyUI queue settings (stored in browser)
  • Smart workflow defaults — Settings maps workflow filenames to model categories automatically
  • Workflow pre-flight — Topics/Variations batch queue validates workflow placeholders before submit
  • ComfyUI param recovery — gallery re-queue restores saved seed/params from queueParams
  • LoRA trigger injection — missing trigger phrases from the LoRA library append on ComfyUI queue
  • LoRA train pipeline — Settings → ComfyUI → LoRA train + Cast flywheel: export keepers to {PROMPT_DATA_DIR}/lora-datasets, durable SQLite jobs, first-party kohya/sd-scripts templates (rank/steps/resolution), optional TRAINER_URL/TRAINER_COMMAND escape hatch, copy weights into COMFYUI_ROOT/models/loras on complete, pin + Prove validation
  • Queue param overrides — optional seed/width/height/cfg/steps overrides in Settings and result panels
  • Queue orchestration panel — home/gallery view of ComfyUI server queue, VRAM, and local tracked jobs
  • Central job queue — /queue page for pending ComfyUI jobs
  • Queue upgrades — /queue shows ComfyUI queue stats (pool totals + per-host depth), failed jobs, and bulk retry
  • Readable queue — jobs lead with where they came from (Day · Evening · Robin, Story · beat 3 “The letter” · retry) and link back; per-job and total time left from recent render times on the same model (also on the active-jobs chip); Run next resubmits a waiting job at the front of ComfyUI's queue (the Day slot / Story beat follows it); jobs queued together group into a batch with Cancel batch
  • Orphan host jobs — Queue lists in-flight Comfy jobs from every pool host that are not in this gallery; Import or Import all to track them
  • VRAM-aware Max → Final — when free VRAM is under ~6 GB, Max queues downgrade to Final
  • Hold Max until idle — optional park for Max jobs until ComfyUI is empty; flush from Orchestration
  • Sampler memory — 4–5★ gallery ratings remember per-model CFG/steps/sampler/scheduler
  • Multi-ComfyUI pool — COMFYUI_POOL plus Settings extras; load-balance skips busy hosts; OOM / dead-host retry fails over; optional sticky preferred host for gallery stills
  • Second-GPU snippet — after Test (or allowlist miss), copy COMFYUI_POOL / COMFYUI_ALLOWED_HOSTS into .env.local; probe never fetches an unallowlisted host
  • Queue artifacts — optional COMFYUI_QUEUE_EXPORT_DIR or Settings overlay writes JSON sidecars after queue
  • Negative A/B — same-seed ComfyUI queue with/without negative for SD-family models
  • Same-seed shootout — queue one prompt across models with identical seed (Settings → Advanced)
  • Auto-improve loop — optional auto-mutate or seed-experiment on high ratings / favorites
  • Prompt recipes — Settings → Advanced chains lint/fix/compact/queue steps
  • Prompt lineage — gallery entries link to Studio history when queued from result panels

Workflows & library

  • Workflow preset packs — import/export bundled presets in Settings workflow library
  • Workflow preset pack builder — add workflows or settings snapshots to packs; install packs into library
  • Workflow diff — Settings compares two workflow JSON files
  • Workflow node auto-map — suggested positive/negative bindings while editing workflow JSON
  • Video prompt builder — /video + local WAN / Hunyuan / LTX, or Fal / Replicate / Grok / Gemini / Runway clips (T2V, I2V, extend). I2V Scan with vision fills Subject and Motion from the first frame. Extend chip: Extend clip (Fal LTX / Grok extensions / Runway Aleph video-to-video), Continue from last frame (Replicate; Fal upload soft-fail), or Stitch continue (Gemini + /api/film/assemble). ChatGPT warns before queue. Compact Video model files Install rows download WAN / Hunyuan Video / LTX weights into COMFYUI_ROOT.
  • Logo / Audio / Mesh (parked Extras) — /logo, /audio, and /mesh remain available under the sidebar Extras group (and ⌘K) but are soft-deprecated specialty tools. Prefer Play + Video for motion; Character page covers pet/fantasy/environment without the legacy /pet /fantasy /background routes (those stay palette aliases).
  • Audio / mesh model files — Settings → ComfyUI → Model assets can still Install Stable Audio Open 1.0 + T5-Base, and Hunyuan3D 2.0 DiT into checkpoints/ / text_encoders/ when using the parked tools.
  • Refine vision scan — /refine Scan with vision fills Current prompt from the reference still (same pattern as Video I2V), then intent + Refine.
  • Vision scan on still tools — Inpaint, Outpaint, Compose (Image 1), ControlNet, and Story From photo share Scan with vision next to Choose from Gallery. Compose also has Scan all images → instruction for multi-slot transfer recipes.
  • Tool-locked model pickers — Video stays on WAN / Hunyuan / LTX; Inpaint / Outpaint / Compose / Refine / From photo stay on edit/img2img. Show all is hidden on video/audio/mesh.
  • Workspace first paint — saved Simple / Play / Studio / Full is read from a cookie so chrome matches on load instead of flashing Simple.
  • Mobile Cut film — /m/story and /m/day Cut film and Save to Cast with the same assembler as desktop (Cast / Gallery deep-links and Campaign complete CTA). Phone is first-class; Desk Continue is optional.
  • ControlNet prompt builder — /controlnet tool for depth/pose/canny/normal/lineart conditioning text
  • ControlNet from image — upload reference for vision-assisted structure extraction on /controlnet
  • ControlNet gallery lineage — gallery → ControlNet handoffs keep parent entry + source image; derivatives filter as ControlNet
  • Multi-ref image prompts — Image tool accepts up to 4 references (/api/image-prompt/multi)
  • IP-Adapter multi-ref merge — vision LLM describes refs and blends into text prompt (no ComfyUI IP-Adapter nodes)
  • Portable IP-Adapter identity — Settings → ComfyUI patches {{IPADAPTER_*}} tokens or auto-inserts IPAdapter chain at queue time

Settings, storage & backup

  • App database (Dexie) — settings, history, presets, workflows, webhooks, and gallery in IndexedDB (comfy-prompt-studio-v1); legacy localStorage migrates on first load
  • Settings cache — target model, detail level, and per-tool options persist across reloads and pages
  • Settings hub — /settings tabs: Overview (Heal & ready, health, backup), LLM, ComfyUI (cluster + assets), Automation, Data, Users (SMTP + invite)
  • Settings env panel — copy .env snippet and re-run LLM/ComfyUI health tests from Overview
  • Heal & ready — one-click first install: system workflows, loader maps, inventory adapt, ComfyUI-Manager pack install on each pool host, restart, health refresh
  • Studio backup v5 — export/import history, settings, gallery, workflows, projects, recipes, and extras (gallery ELO, views, appearance, shortcuts). v1–v4 still import
  • Overview one-click backup — Export / Import on Settings → Overview for a new browser or machine
  • Full user backup — Profile downloads/restores the same studio backup JSON
  • Backup reminder — Overview/Data warn when no recent export
  • Prompt sidecar — download JSON sidecar (prompt, model, diagnostics, seed) from result panels or Studio history
  • Sidecar import — load sidecar JSON on Gallery, Lint, and Variations to restore prompts or re-queue
  • Server storage sync — optional PROMPT_DATA_DIR SQLite (studio.sqlite) via /api/storage; per-user scopes when logged in
  • Server storage pull — Settings advanced panel restores server namespaces into the app database
  • Auto storage sync — pull on login when browser is empty; conflict merge UI when local/server diverge
  • Auto-push storage — history and gallery saves debounce-push to server when storage sync is enabled
  • Multi-tab sync — BroadcastChannel refreshes gallery/history across open tabs
  • Encrypted exports — POST /api/storage/export with optional passphrase
  • Encrypted server export — Settings → Advanced exports signed-in user data with optional passphrase
  • Auto-save on queue — Settings toggle; skips duplicate history when you already saved manually

Auth, admin & API

  • User accounts & feature ACL — optional login, groups, blocked features, viewer role, per-user ComfyUI URL override, API quotas (Settings → Users)
  • Invite by email — admin creates a user (or re-sends) via SMTP; 1-hour /login?reset= link. Password optional on invite
  • SMTP overlay — Settings → Users persists host/port/from in SQLite when PROMPT_DATA_DIR is set; Send test uses the overlay immediately
  • Password reset — /login forgot-password + POST /api/auth/reset-password; same mailer as invites
  • Profile — password change, export toggle, scheduled campaign settings, 2FA, sessions, email, appearance
  • Admin tools — audit log, user impersonation, shared read-only preset library, analytics trends over time
  • Per-user API keys — pt_… tokens for CLI/inbound hooks with user quotas
  • Inbound webhooks — POST /api/hooks/generate with user API key or INBOUND_WEBHOOK_SECRET
  • TOTP 2FA — optional authenticator setup on Profile
  • Session management — list/revoke sessions on Profile
  • API usage & rate limits — proxy logs usage; optional PROMPT_API_TOKEN + rate limit env vars
  • LLM usage dashboard — per-user call/token stats in Settings → Advanced
  • Observability dashboard — Settings advanced panel shows API volume, errors, and slow routes
  • Shared projects — admin assigns group-scoped projects via /api/shared-projects
  • Shared projects UI — admin CRUD in Settings → Users; adopt in Studio → Projects
  • Plugin registry — /plugins Sidebar bookmarks: custom localStorage nav bookmarks (href-only; not a runnable plugin runtime; cap 32)
  • Plugin iframe host — /plugins Runtime plugins: installable manifests embed tools at /plugins/[id] with queue / apply-prompt / apply-model / apply-quality / apply-engine / apply-lora-stack / patch-workflow-tokens / write-gallery-tag / pick-gallery (docs/plugin-iframe-host.md). Choose runtime when you need iframe or queue hooks; choose bookmarks for simple deep links.
  • Server plugin registry — {PROMPT_DATA_DIR}/plugins via /api/plugins/server (ZIP/URL install, optional PROMPT_PLUGIN_HMAC_SECRET); privileged queue-preflight / queue-post hooks inside the Comfy queue path with allowlisted prompt/workflow rewrite; sync into the UI runtime list. Auth on → plugins feature + admin for install/remove.
  • Queue failure playbooks — missing-node and loader failures deep-link toast CTAs to Settings sections
  • Simple first-success — first completed render in Simple mode advances onboarding and nudges Gallery review; checklist offers Open Play workspace from first-queue-success onward (earlier than campaign-only), and Simple Essentials includes /play
  • CLI — npm run prompt:cli -- duo --hints "..." over the HTTP API

Automation & integrations

  • Quieter Settings — Essentials ComfyUI shows engine / connection / assets / queue; maps & patching live under Show all with Reliable / Studio / Raw queue profiles and collapsed Expert maps; Cast-active IP-Adapter soft-hides

  • Automation hub (Settings → Automation) — webhooks, avoided tokens, browser/server scheduled batch; links to ComfyUI notifications and Profile email campaigns

  • Completion notifications — optional browser notifications when ComfyUI jobs finish (Settings)

  • Notification center — in-app alerts bell in sidebar when jobs complete
  • Webhooks — POST queue, prompt, and session events to an external URL via server proxy (comfyui.job.*, prompt.generated, prompt.history.saved, session.recipe.saved, scheduled batch)
  • Webhook event log — Settings shows recent webhook deliveries with retry
  • Webhook log UI — event filter and payload preview on Settings
  • Webhook templates — Discord/Slack rich payload formats in Settings
  • Scheduled batch — Settings configures periodic random-scene/topics generation (+ optional ComfyUI queue)
  • Server scheduled batch — SERVER_SCHEDULED_BATCH=true or manual POST /api/scheduled-batch/run
  • Best-of-N campaigns — scheduled profile or Automation tab runs optionally over-generate (2–4×) and LLM-rank prompts by text quality before queue; with Vision-rank enabled, queues all variants then vision-scores outputs and culls losers from the gallery; Profile server campaigns and headless server batch support the same path; /api/best-of-n/rank-images ranks generated images with LLM_VISION_MODEL
  • Prompt recipes API — /api/recipes/run executes lint/fix/compact/queue chains server-side; result panels call it from Recipes & shootout shortcuts
  • Email notifications — SMTP alerts for invites, password reset, batch/campaign completion, and password changes (Settings → Users → SMTP; Profile → Email)
  • Docker Compose — docker/: copy .env.example to .env, docker compose up -d --build; data in a volume; optional Ollama / ComfyUI profiles
  • GitHub Actions CI — runs unit tests, build, and Playwright smoke on push/PR

UI & UX

  • Workspace levels — Film (the default: Cast, Look & Outfit, Day / Story, Gallery and Cut), Studio (more control: prompts, references, model settings and every tool, with the film in progress kept at the top of the sidebar) and, inside Studio, More tools (specialist tools, workflow graphs, integrations); switch in Profile → Appearance (a saved Simple opens Film, a saved Full opens Studio); the saved mode is applied on first paint
  • Unified first-run path — Welcome goal chooser → Heal & ready → landing CTA; Settings Heal and first-queue modal share dismiss state; Generate offers Photoreal / Illustration / Edit / Video goal chips before the full model list
  • Post-Heal checklist — Connection first-run card lists Generate → Queue → Gallery review after Heal succeeds
  • Active jobs chip — sidebar and mobile header show live queue count with a deep link to /queue
  • Mobile Studio offer — narrow viewports get a dismissible banner to open /m
  • Lean gallery discovery — tip strip teaches Select → Compare / Collect / Group when nothing is selected
  • Gallery group management — Groups rail plus rename/delete for custom groups
  • Semantic search inline — Match chip sits next to gallery search (non-lean)
  • First-success tray nudge — first lean-mode completed render uses notification + celebrate even when toasts are muted
  • Visual polish — restrained button/chip motion (no lift/glow), solid docks instead of stacked glass bars, quieter gallery overlays and menu labels, ToolBadge without pill chrome, Settings jump via select. Default (comfortable) density tightened toward studio tooling; Film active steps use real accent tokens; Dashboard elevates Play as the primary path when not resuming a draft.
  • First-run queue deep link — post-Heal checklist and /?source=random&autogen=1&autoqueue=1 auto-generate + queue a random scene
  • Dashboard output actions — Re-queue, Refine, Story, Edit, Hints on recent stills
  • Command palette context — active project, recent gallery outputs, Heal & ready action
  • Mobile gallery filters — layout/density/rating as selects on narrow viewports
  • Home dashboard — pending ComfyUI jobs, recent outputs, and active project on /dashboard
  • Onboarding checklist — Dashboard getting-started steps
  • Command palette — Ctrl+K / ⌘K quick navigation across tools
  • Keyboard shortcuts — Ctrl+Enter generate, Ctrl+Shift+C copy pair, Ctrl+Shift+G queue ComfyUI; / focuses scene preset search
  • Keyboard shortcut editor — customize bindings on Profile
  • Light theme — Profile → Appearance switches Auto / Light / Dark
  • Ambient background — subtle animated orbs; intensity toggle on Profile → Appearance
  • UI density & calm mode — compact spacing and reduced motion on Profile → Appearance