Play film guide¶

The short version

This page is the full reference. For quick, task-by-task guides see Your first film, Day, Story, Cut a film and Finishing tools. The Look page (step 2 below) has been retired: Film goes Cast → Outfit → Day, and /moodboard opens the Cast's looks.

The Play workspace is a guided film loop on one Cast character: Look → Outfit → Day → Cut film (Story optional after the first cut) → Save to Cast. This page is the product walkthrough; ops and env vars live in the operator guide and configuration.

Jump to: When to use Play · Step-by-step · Dashboard metrics · Share & resume · Play chrome & habits · Mobile vs desk

When to use Play¶

Workspace	Best for
Play (default)	One character, one look, one day-in-the-life film
Simple	Essentials — Generate, Gallery, Queue, Cast
Studio / Full	History, compare, templates, advanced queue controls

Switch modes from Profile → Appearance. Play slimmed chrome hides draft-preview noise and optional Story until you cut your first film.

First launch asks What do you want to make? (Character / Scene·Film / Image / Surprise). Every goal opens in Film; Studio (Profile → Appearance) adds control, with your film kept at the top of its sidebar. The welcome says up front whether ComfyUI is ready. On a phone, use Mobile Studio (/m) — the same Film · Day · Outfit · Story tabs as the desk, with Look, Cast, Gallery and Queue under More.

First film in minutes

Make a starter film seeds a Cast lead, skips Look/Outfit, opens Day with auto-queue, and lands you on Cut. If Comfy is offline, tap Use demo stills to practice the cut. Welcome shows a sample morning→night reel so you know what you're making.

Step-by-step¶

1. Open Film (`/play` · `/m/film`)¶

Create or pick a Cast character (name + optional From photo first; Part / look traits under More traits) → Create & continue to Look, or tap Make a starter film / Make one like this under the sample reel. Phone Film remaps every handoff onto /m/... so you stay in Mobile Studio.

Resume — one Continue card mid-film; Restart at Look stays secondary; Jump to starter Day is under More.
Mission control — funnel chips + Identity ready; full Steps under Edit when mid-film or complete.
Film complete — Watch on Cast · Same look, new Day · Continue in Story · Start new film.
Persistence triad is compact; habit nudge is desk-only on the hub (phone shell already owns it).

Deep links: /play?character=<id> · /m/film?character=<id>.

2. Look (`/moodboard`)¶

Stack reference tiles (mood, lighting, location, style, palette). Optional gallery stills per tile.

Phase strip — Tiles → Extract → Plate → Continue; status line shows tiles · plate · pack and why Extract / Queue is blocked.
Presets — one row; pick a preset, then Load into board or Use for today — skip Outfit.
Tile board — every tile as a thumbnail (image, role, label) plus + Add tile; tap one to edit it. Drop images onto the board or paste one (Ctrl/⌘+V) and each becomes a tile (up to four). With a vision model set up, a new image's role is suggested (mood / lighting / location / style / palette) — never over a role you picked.
Extract look is the film path; Preview prompt (text only) and Queue scene (an optional still) sit under More.
Extract look — builds a session look pack (vision merge when tiles have images), then queues a full-body Outfit plate in minimal base clothing so try-on / Day have a clean body ref. Soft-advance counts down to Outfit with Go to Day instead.
Skip look · Day (desk + phone) — jump to Day without extracting when you already have a Cast lead.
Look plate — upload / Gallery / Remove on Look (desk + phone); after a queued scene still, Keep as plate or requeue.
Continue to Outfit / Day — hand off vibe notes + optional wardrobe lock.
Save on Cast / Export JSON / Share this look — under More (power users).

Deep link: /moodboard?character=<id>.

3. Outfit (`/fitting`)¶

Lock a character plate, browse wardrobe kits, queue try-ons.

Get started — with no Cast lead, a card at the top offers Make a starter film or jumps to the Character picker. With no plate, the Plate section offers Upload plate, Choose from Gallery or Open Look (extract a look → new plate).
Phase strip — Plate → Try-on → Keep → Day; status line shows plate · kit/BYO and why Queue is blocked.
Clothing — one row says what she wears (the kit or your photo, and the shoes); Choose… opens the Clothing sheet, the same picker as Day and Story. In it, Kits — Clothing type quick picks (All, Full outfits, Tops, Bottoms, Outerwear, Formal, Swimwear; More types… for the rest). The picked kit shows large with its name and position; Prev / Next appear once a kit is picked.
Preview vs Queue — draft thumbs vs full-quality try-on (explained under Draft previews at the bottom of the Clothing sheet, collapsed by default).
Quality — Fast / Balanced / Best, under Clothing. Fast = Good render, front only, no review; Balanced (the default) = Good render + Front and back; Best = Best render + Front and back + Auto-review try-ons. Advanced holds the two switches, Good / Best by hand and the try-on notes; changing a switch there shows the preset as Custom. The Engine panel says Set by Quality: Balanced.
Keep on each compare card and in the lightbox; Pass and Requeue · new seed in the card's ⋯ menu (and in the lightbox) — Pass dismisses a try-on; Skip kit advances the wardrobe deck (off until a kit is picked). Tap a picture for full size.
Auto-review try-ons (the Best preset, or the switch under Advanced) — each try-on in Compare gets a face match against the plate (needs ComfyUI_FaceAnalysis) and a vision read of the outfit, face and hands (needs a vision LLM): the card shows Face 64% · Outfit 4/5, warns on a drifted face, an outfit that doesn't match the kit, or broken hands, and marks the best clean try-on Best match. Scores only — nothing is requeued. Face scores feed Face match by model on the Dashboard. A missing node pack or vision model turns that check off with a note.
BYO clothing — Upload worn photo (someone wearing it; the clothes are extracted to a packshot) or Upload packshot (clothes on a plain background, used as is); Save for later keeps it in durable browser storage (desk + phone).
Footwear — in the Clothing sheet, pick shoes the way you pick clothing: Kits (36 with packshots — sneakers, heels, boots, flats, sandals, at home — plus Auto and Barefoot), My shoes (a worn photo: vision reads the shoes and they are cut out into a packshot; or a ready packshot), or In words. The shoes are named on their own line in every clothed still; on Qwen Edit 2511 the packshot also rides in Image 2 under the clothing (on Rapid AIO the picture let the dress drift, so there the words do the work). A worn photo needs a vision model. The same picker is on Day and Story. Save for later keeps a shoe photo under Saved shoes so you can re-pick it in any session; take a pair off before removing it from the list. Browse opens every catalog pair (or every saved pair) in a searchable grid, like clothing. A worn photo is turned into an empty-shoe product shot — sandals and strappy heels included.
Pose — As the plate keeps the plate's stance, Custom pose opens the pose editor, and From a photo reads the pose of someone in a photo (only the pose is used). In the editor drag a hand or foot and the elbow or knee follows; drag the body to move it; turn, tilt and spin by dragging; Bend forward / back bends at the waist. Start from Stand / Sit / Kneel / Lie down / Walk or A Day pose… (any of Day's one-person poses), and Save to My poses. The box The prompt will open with shows the pose in words — that line leads the try-on prompt, because the models follow the words over the pose map; check it before you queue. In the editor, Quick positions set the arms or legs in one tap (both sides or one), Copy a side mirrors a limb you posed by hand, and More poses… shows Day's named poses as small figures to start from. Show her picture puts the Cast's picture faintly behind the figure, Show where it started leaves a dashed outline of the pose you opened with, and the Head row turns the head (the figure has eyes, and a head turned to the side is drawn in profile).
Character — the Look picker is labelled; saving or renaming a look sits under Save or rename this look…. Forget asks before deleting the Cast record.
Continue to Day — primary CTA after a keeper. Skip outfit · Day (desk + phone) if you want stills without a kit lock.
Save kit to Cast — under More. Phone can upload / pick a Gallery plate in-place.

Deep links: /fitting?character=<id>&wardrobe=<kit>.

4. Day (`/day`)¶

Four slots (Morning → Night) by default, with wardrobe, setting, and beat per slot. The page is a board: a one-line plan bar above the cards (Stills · Mood · People · Weather · Quality · Advanced), the slot cards, one action row, then the Day reel. Everything about one slot opens in its slot sheet (a right drawer on desk, a bottom sheet on a phone); Setup (Cast lead, look, plate) is the status chip beside the title. The Stills segments in the plan bar set the Day's length — 2, 3, 4, 6 or 8 stills; longer Days add Late morning / Late afternoon / Late evening / Late night slots with their own later-hours activities — brunch, markets and errands; golden-hour walks and picnics; dinner, drinks and a show; after-hours at home — and matching Vacation scenes (harbor brunch, sunset cruise, rooftop dancing, night market) and Sport windows (club matches, after-work leagues, late training). The heat moods get later-hours pools too — lazy late-morning, golden-hour siesta, back from dinner, 3 a.m. — with the same Solo / Duo split and indoor settings as their daypart pools. Every slot keeps its plan when you change the length.

Suggest day fills a morning→night Setting / Beat plan you can edit; Queue day keeps your plan (only fills blanks). Per-slot Reroll plan refreshes one card without re-queuing.
Day clips run on LTX-2.5 by default, whatever Video → Model says — faster, her face stayed closer to the still, and an event's joins vanished on it. Two-person adult clips stay on WAN unless you tick Use LTX-2.5 for two-person adult clips too. To use WAN instead, set Day clips on → WAN in the engine note under the cards (talking clips stay on LTX-2.5 — only it makes sound).
One event — under the cards, switch the idea row from Whole day to One event and write one line (a dinner date at a candlelit trattoria). The LLM writes every slot as a consecutive moment of that event in one place and one time of day, each with the move into the next; every card is labelled with the event's time. Queue day makes the first moment the usual Day way, then each later moment as a same-camera edit of the moment before (Edit 2511), one after another — so the room, light, camera and clothes carry over. Animate ends each clip on the next moment's still (an end pose — one you pick by hand still wins), so in Cut the clips run straight into each other as one video. She holds what she uses (one or two props) from first to last and keeps doing something with it to the very end — no picking up or putting down, no winding down — and keeps one posture; a moment with a spoken line says it while moving into the next moment, and the last clip moves into the plan's ending (or ends on its own still when that edit drifted). Four moments give the cleanest film. The moments name nobody else (an edit paints a mentioned waiter or passer-by in) and stay in the picture. Rough moments happen — re-roll that card's still or clip. Back to a whole day puts the Day's own scenes and time-of-day labels back; a new Day or Tomorrow is a whole day again. Needs the LLM (Settings → LLM); clothed moods only.
Slot sheet — tap a card's name or Edit (or ⋯ → Edit slot, or a slot with nothing rendered yet): Setting and Beat, the pose preview, the look, a Clothing row that opens the picker (kit deck, your photo, footwear) as its own sheet, Setting presets, takes (Redo · same seed), End pose, and Queue / Animate this slot. ‹ › in its header walks the slots; Escape or the backdrop closes it. A finished card's picture opens the still full size instead; the card's one ⋯ menu holds Edit slot, Open full size, Requeue, Animate and Open in ComfyUI (an unrendered card: Queue this slot only, Reroll plan).
Quality (plan bar) — one preset that sets the render quality (the Engine's Good / Best for Day stills) and the checks: Fast = Good render, no checks; Balanced (default) = Best render + Face finish + Redo pose misses once + Pick the best engine per pose; Best adds Auto-review stills, Best of two for hard poses and Face boost (needs a vision model and DWPose). Advanced ▾ opens the seven switches (plus Good / Best by hand, the pose pack and Day notes); change one and the preset reads Custom until you pick a preset again. Pose over plate and the adult-appearance gate are not part of the preset. Days from before read their preset from the switches they had.
Queue day is disabled with a one-line reason when Cast, look plate, or isolate-on-white isn’t ready.
Slot progress — a queued slot shows its queue position (#3 in queue, Next in queue) and, once ComfyUI starts it, Rendering · 45% with a progress bar. Retry N flagged (next to Queue day, with Auto-review on) requeues every flagged still and clip at once.
Plate from Day — the Setup sheet (the Nora · plate ready chip beside the title) can upload a photo or pick one from Gallery; it becomes the Cast's look plate, so Outfit and Story use the same one.
Cut — the Ready to cut banner keeps Cut film / Queue rest up front, lists what the cut would quietly do (a slot left out because its still failed or is still rendering, a failed clip replaced by its still, a line with no talking clip) and says which checks have run (Checked: faces 3 of 4 · Not checked: pose); crossfade, shape and size, slow zoom, titles, audio bed and the Music / Lines levels sit under Cut options. Above the clothing row, Held across the Day lists what every still keeps (face from the plate, the Day's clothes, the Cast's places, the weather); under the buttons Day says how long Queue the rest / Animate all would take here.
Outfit kit + BYO — same clothing strip as Outfit (Image 2); optional pose stick-figure on Image 3 so Edit can unlock stance (Image 3 Edit only — pose guides do not attach ControlNet). Everyday beats map to stretch, wave, drink, carry, read, rail, pockets, cross-legged on the floor, lounging on elbows, lying on the front or side, perched on an edge, hands behind the head, arms up, selfie, camera to the eye, cooking, laptop, eating, and more — not only stand/walk/sit. With People set to Mixed or Duo, companion beats also draw holding hands, piggyback, high five, toasting, head on a shoulder, and a selfie together. Sport maps beat text to mid-action athletic Image 3 layouts (sprint, yoga, cycle, swing, jump shot, kick, lunge, handstand, swim, spike, box, surf, squat, deadlift, push-up, plank, pull-up, skate, etc.). Solo stills stay one adult by default; set People (plan bar) to Mixed or Duo for friend/selfie second adults — the Partner tiles then appear under the bar. Everyday / Suggestive / Sport / Vacation / Intimate / Raunchy sit in the plan bar's Mood list (Kind of day / Heat) (Intimate + Raunchy need NSFW env; Sport picks from 23 sports (including gym strength training and skateboarding) + cycling road/gravel/MTB/CX/track disciplines with mid-action poses for that time of day; Vacation picks matched travel poses + venues — hotel, pool, market, balcony, rooftop — and keeps Outfit Keep on). Under Intimate or Raunchy, pick Mixed / Solo / Duo so heat isn’t duo-only. Each slot card shows Setting · Beat; an unplanned slot shows + Add a beat, which opens its sheet. With no Cast lead or plate yet, a get-started card at the top of Day offers a starter film or opens Setup; pose and camera rotate with the scene. In the slot sheet, under Setting · Beat, the pose preview shows the pose guide as a stick figure with its name — Change pose overrides what the beat matched, Try another redraws a variant. Camera sets the angle (auto, front, side, overhead, low) and two-person poses can put the lead on the left or right. From a photo… reads the pose from any photo (DWPose in ComfyUI; OpenPose guide styles) and draws that exact skeleton, with the photo shown beside the figure — a one-person slot uses the largest person in it, a two-person slot needs both people in frame, and a photo with nobody found says so; Save to pose library keeps it. After an Auto-review pose miss, the editor shows the still's skeleton over the guide and names the limbs that were off. Edit joints lets you drag the figure into shape; Look sets where the Cast looks (camera, away, down, at the other person). When Auto-review runs out of rerolls it keeps the best attempt, not the last. Suggest day avoids repeating the same gesture across slots. Pose pack (under Advanced) poses the whole Day from a theme — Fitness, Dance, Portrait, Lounging, Street style, Beach — or from one of My packs (Save as pack keeps the Day's slot poses); Fill empty beats writes a matching beat only where a slot has none, and Clear poses puts every slot back to the pose from the beat.
Footwear — the Footwear picker in the slot sheet's Clothing sheet (see Outfit) applies to clothed stills. Sport keeps the sport's own shoes, nude stills skip it, and a beat that is about the feet (heels in one hand, barefoot on the sand) keeps its own say.
Dress plate — when you picked clothing or shoes, the first Queue dresses the Cast once (one extra render, about a minute; the status line says Dressing … first) and every clothed still then starts from that dressed plate. Day and Story share it — one plate per Cast plate + clothing + shoes — and a try-on you Keep in Outfit becomes that plate, so nothing is rendered twice. It is kept until you change the Cast plate, the clothing, the shoes or the engine, so later sessions go straight to the stills. The plate is shown under the status line: click it for a large view, or Dress her again if it came out wrong. With shoes picked and a vision model set up, the plate's feet are checked when it lands; a barefoot or wrongly shod plate gets one feet pass on Edit 2511 (The dress plate came out without the picked shoes — putting them on…) and the stills start from the shod plate. On Qwen-Image-Edit-2511 the dressed plate is the still's starting image; on Qwen Rapid AIO and Qwen-Image 2.1 it takes the clothing image's place (they identify her from a face crop). Other engines dress her per still as before. Intimate / Raunchy, adult-rated stories and Sport do not use it.
Prompt check — when you queue, each still's prompt is checked for contradictions that would otherwise only show as a bad render (a one-person still that describes two people, shoes on a barefoot or swimming scene, an image the prompt mentions that is not attached). A notice names the still; it is queued either way. Story does the same.
Which engine for Day stills — from a sweep of every Day pose on one Cast (October 2026, 24 GB card; 41 one-person Everyday poses, 9 two-person, 30 Sport):

Engine	A still takes	One-person	Two-person	Sport	Know before you pick it
Qwen-Image-Edit-2511 Lightning (8-step)	about 55 s	40 of 41	9 of 9	28 of 30	The liveliest stills — action, props and angles. On two-person stills the lead's outfit is close, not exact. Adult nude stills are handed to Rapid AIO NSFW. Takes the footwear picture
Qwen Rapid AIO (Edit)	about 20 s	40–41 of 41	9 of 9	26–27 of 29	Fastest; keeps the exact outfit on two-person stills. Likeness is looser on action shots — turn on Face finish when it matters
Qwen-Image 2.1 — Turbo (8-step)	about 20 s	23 of 24	about 15½ of 16	not swept yet	The Qwen team's own 8-step: sharper than Pruna and clean lying down (5 of 5 where Pruna melted or drew a twin on 3 of 5). Qwen Research License — not for commercial use by default
Qwen-Image 2.1 — Pruna (8-step)	about 20 s	40 of 41	handed to Rapid AIO	26 of 29	The closest likeness of the three, and it follows a pose you drew in Outfit exactly. Everyday stills are front-facing and dimly lit; lying on one side can ghost

Counts are stills that showed their beat, judged by eye.

On two-person stills the lead's likeness drifts more than on one-person stills (the faces are smaller). Face finish repairs only her face there, never the partner's, and keeps the result only when it is measurably closer.

Face finish first measures her face (a few seconds): a face already close to the Cast face (distance 0.35 or less) is left as it is (Face finish skipped — her face is already close), and a finished face is kept only when it came out closer — on one person too. On Balanced the pose check (and its gesture / realism question) runs on each still as it lands and Face finish waits for it, so a still that is redone for its pose is never face-finished first. The quick checks (pose, face, cut-outs) run right after the job ComfyUI is on, ahead of queued renders. - Engine — the Engine chip in the header opens model & workflow; its Quality reads Set by Quality: Balanced on Day (the preset owns Good / Best — override it under Advanced). Plate Day/Story stills need an Edit model: Qwen Rapid AIO (Edit) defaults to Phr00t SFW v23; pick Qwen Rapid AIO (Edit NSFW) for the NSFW v23 merge (Intimate / Raunchy). Intimate/Raunchy nude Day and Story queues auto-snap Rapid AIO → Edit NSFW (required for bare-skin NSFW — SFW will not deliver that). Nude Day auto-crops a ≥1.1MP face window when face lock duplicates lingerie; edit lead uses clear natural language: indoor SETTING first (no beach/sand), then “clothes are now gone” / bare skin / zero fabric, then the beat/Image 3 pose named explicitly — do not name bra/panties/beige in the positive (those bans stay in the Rapid negative pack); Rapid stays 4–8 steps / CFG 1 / euler_a+simple (Phr00t AIO — never raise CFG); Solo nude IP lock caps at 0.04. Map to v21 in Settings if preferred. Edit-2511 Lightning remains the default snap from Qwen 2512 T2I. - Auto-review stills (on with the Best preset, or by hand under Advanced) — each finished still is vision-checked (face, hands, outfit); a broken slot is requeued with a targeted fix up to two times, then flagged for you to retry or reroll. Needs a vision LLM (LLM_VISION_MODEL, a Settings → LLM vision model, or one found on the LLM server — Settings → Overview → Play checks says which). It reviews one still at a time and only while the queue is idle. With comfyui_controlnet_aux (DWPose) in ComfyUI it also checks the still followed its Image 3 pose guide, and with ComfyUI_FaceAnalysis it measures whether a solo still's face matches the plate — a clear pose miss or a different person is requeued like any other defect, and the review line shows pose match 82% · face match 64%. Missing node packs just turn those checks off with a note. Animate clips get a lighter check once they land: frames are sampled in the browser, and a clip that barely moves, has blank frames, or whose face drifted from the plate by its last frame is flagged on the slot card (Clip: face drifted from the Cast (18%) — try Animate again) — never re-animated automatically. Pass rate, Pose match by guide, Pose match by layout and Face match by model (with an Animate clips row) show on the Dashboard. A layout that keeps missing (8+ checks averaging under 45%) is first spelled out in words in the prompt; only if it still misses with the words (4+ checks) is it routed around — a saved pose-library skeleton if you have one, else the plain posture — unless you picked it. A pose-miss retry always adds the words and names the limbs to fix. - When a Day plate is set and the still is solo, the review also gets an identity pair — your plate on the left, the new still on the right — and scores whether it still looks like your Cast. A mismatch only warns ("Passed · worth a look · face may not match the Cast"); it never spends a reroll, because a small face in a wide shot is a weak signal. Duo / companion stills skip the pair, since the reviewer can't tell which face to match. - Each slot card shows its review result — Check this still (with the reason) when the gate gave up after two rerolls, or a quiet Passed after N rerolls note when a requeue fixed it. Use the card's ⋯ menu — Requeue or Reroll plan — from there. - Animate → Cut — after stills land, Animate all becomes the action row's primary (clips preferred), then Cut in the banner; Cut still works from stills alone. One primary per phase: Queue day → Animate all → Cut → Save. - Cut film — server ffmpeg when available, browser MediaRecorder fallback if the server encode fails; optional crossfade, Shape (Widescreen 16:9, Vertical 9:16 for Shorts / Reels / Stories, Square 1:1 for feed posts — vertical and square crop to fill) and Size (720p or 1080p), and audio bed (Upload audio or paste a URL). Slow zoom on stills (on by default) gives each still a gentle push-in or pull-out so a stills-only cut doesn't read as a slideshow. Title & captions opens the cut with a title card (the Cast's name over Season N · Episode M on Day) and fades a short caption — the slot or beat title — over the first seconds of each shot; Save poster then carries the same title. - After cut: celebrate with Watch / Save on Cast (manual — no auto-advance). Story is optional. - Remix — next to Same look, new Day: Same Day, new outfit keeps every Setting and Beat, clears the stills and kits, and sends you to Outfit to pick a fresh Keep; Themed day… re-runs the same look with a themed Setting + Beat set (Rainy day, Weekend out, Workday, Cozy home, City trip — everyday mood). Deep link: /day?character=<id>&remix=1&theme=<id>. - Films → What's in it · Re-cut (Cast page) — a film cut in 2.6 or later lists its shots and re-cuts at another shape / size as a new film. - Undo / Redo (Day, beside the plan) — steps back through plan edits (places, beats, outfits, lines, poses, slot count). - Seasons → What went into it / Open this Day again (Cast page) — every Day cut keeps its Day on the episode: its slots' stills and clips, places, beats and lines. Open this Day again puts it back on Day; the Day that was there is kept aside behind Bring back the Day you had. - Save poster (also on Story and Cast film) — renders a poster/thumbnail frame from a finished still (the slot you are editing, else the first finished one), center-cropped to the cut's aspect (so a vertical cut gets a vertical poster), saved to Gallery next to the film and downloaded. - In the Gallery — Day, Outfit and Look stills open back in their own tool on their Cast (they used to open Generate). A Cast filter row (look-plate thumb + name) narrows to one Cast, and with a Cast picked a Look row narrows to the look the stills were made in; each Cast page has See all in Gallery. A finished still's menu has Open in ComfyUI (the exact graph, saved to ComfyUI's Workflows). Stills checked by Auto-review or Story's pose check carry a pose 82% · face 64% badge (warning-tinted on a miss), with a Missed pose / face filter and a Best pose / face match sort. Use this pose… (card menu) reads a still's skeleton with DWPose and saves it to the pose library or sets it as a Day slot's or Story beat's pose. - Cut options — Length (as the shots hold · Fit to the music · 15 / 30 / 60 s) and Cut on the beat (with a music track), plus a shot list: reorder, leave out, edit captions (Day captions default to the beat), set still holds, frame stills (waist up / chest up: a crop around her face), trim clips (from … to …: only that piece plays, its voice trimmed with it), pick the Poster (a still, or a clip at a chosen second), and a face mark per shot (Looks like Nora / Face may drift / Not Nora?). Before encoding, Cut checks the shots' files (Checking the shots…: missing, black, a talking clip with no sound) and lists them with any shot Auto-review flagged or that missed its pose / face — Retry them first, Leave them out or Cut anyway. Story Cut works the same way. Each finished Day still carries the same face mark on its card. - Season — each Cut that lands in Gallery is recorded as an episode of that Cast's current Season (Robin · Season 1). With two or more episodes, Stitch season joins them oldest-first into one reel saved to Gallery; New season closes the current one so the next film starts Season 2.

Deep links: /day?character=<id>&wardrobe=<kit> · Look handoff: ?from=look.

What sound needs (Settings → ComfyUI → Models, filter LTX-2.5; Settings → Overview → Play checks shows what is missing):
Talking clips, conversations, Add voice: the five LTX-2.5 files (distilled int8 transformer, Gemma 4 text encoder, video VAE, audio VAE, latent upscaler — about 39 GB). Lightricks gates them: accept the terms on huggingface.co/Lightricks/LTX-2.5 once, then set HF_TOKEN.
Kept Cast voice (optional): the LTX-2.3 ID-LoRA.
Face pass (recommended): ComfyUI-ReActor (Install in Play checks) and CodeFormer face restore.
Score this film: ACE-Step 1.5 (film scores), 10 GB, on ComfyUI's built-in audio nodes.
Room sound under talking clips: castcut-nodes 1.7.0 or later in ComfyUI.
Refine detail (optional): ltx-2.5-22b-ic-lora-refine-details-1.0.safetensors from huggingface.co/Lightricks/LTX-2.5-22b-IC-LoRA-Refine-Details (accept its license first) and a working ComfyUI-LTXVideo node pack.
Animate — the slot's sheet (or a card's ⋯ → Animate) opens what the clip will do: the default direction from the beat, editable. Write your own and it is formatted for the engine the clip runs on.
Clip engines — with LTX-2.5 picked in Video, every clip runs on LTX-2.5 except two-person adult clips, which stay on WAN unless Use LTX-2.5 for two-person adult clips too is ticked under Animate. Clothed one-person clips can be sent back to WAN (untick Use LTX-2.5 for clothed one-person clips). LTX holds the still harder on one-person and adult clips — her face stays closer to the still than on WAN. Refine detail gives every LTX clip one more pass for crisper texture (about twice as long; the face can shift a little).
Talking clips — a slot (or Story scene) with a Line in the clip animates on LTX-2.5 with sound: the Cast says the line, lip-synced (about 5 s). How it's said picks the delivery (Natural, Whisper, Laughing, Excited, Tender, Teasing, Angry, Nervous, Scared, Shouting, Sing-song). A full-body still stays a wide shot and she keeps her pose — standing or sitting; around her the scene keeps moving (steam from a coffee, a fountain, water, leaves, people passing — whatever the place and the line name), and the place's room sound plays quietly under her voice (Room sound under the voice, under the line; needs castcut-nodes 1.7.0). Suggest a line writes one from the beat, the time of day, the mood and the Cast's personality (Intimate / Raunchy lines sound like the moment, not small talk). Two-person adult clips on WAN make no sound — Add voice (⋯, or automatically for a slot with a line) gives the finished clip LTX's soundtrack with the picture untouched, and Add sound gives any silent clip the sound of its place. Cut film keeps the speech and turns the music down under it.
Make it 30 s (a finished clip's ⋯ menu, Day and Story) — the clip carries on to about 30 seconds in parts that continue from each other (a few minutes). In the sheet, say where it should go and Write the beats (one line per part, each editable; add parts for a longer clip). Small steps in the same place work best; the face and clothes can drift late in the clip.
Conversations — on a two-person (clothed) still, add Their reply under the line: the clip has the lead say the line and the other person answer, each in their own voice. Suggest a reply writes the answer.
Score this film (Cut options) — an original instrumental in the Day mood / Story tone, sized to the cut (ACE-Step 1.5, Settings → ComfyUI → Models).
Voice (Cast → Bible) — Audition voices renders four short clips of the Cast saying hello in different voices; keep one (or Use this voice on any talking clip, or Use my own recording / Record for a real voice — about 5 s of one person talking) and nudge it Deeper / Higher. Later talking clips are steered toward it with the LTX ID-LoRA (Settings → ComfyUI → Models).
Places (Cast → Bible) — every Cast has one home and one workplace; any still set in one of their rooms (bedroom, kitchen, living room, bathroom, hallway, balcony; desk, meeting room, office kitchen) uses that place's look, so the rooms match from slot to slot. Pick another design, describe your own, or turn it off. Clothed moods and Story; not Intimate / Raunchy yet.

5. Story (`/story`)¶

Optional after the first Day cut: continues the Cast lead you started on Film (no re-casting).

Needs an active Cast character — empty Story sends you to Film / Cast (Part and From photo are set there).
Beat picker first (desk + phone): Roll four scenes leads the card; Tone / Content / Setting / notes (and adult Solo / Duo / Mixed) fold under Story settings · Silly · PG-13 · any setting; Roll shows a clear block reason when bible/plate is missing.
Steer the story yourself — under the four cards, Or write what happens next plays a scene you type (at the end, your own ending). Take back the last scene removes only the last scene; Start the story over clears the reel and uses the current Story settings from the first scene (settings changed mid-story otherwise apply from the next scene on); it asks whether to Keep the bible or Write a new bible for the same lead. Stills and clips already made stay in the Gallery.
One ⋯ per scene card — Edit scene, Pose…, Open full size, Queue still, Animate, Play another still / clip, Extend, Copy prompt and Open in ComfyUI live in each reel card's ⋯ menu; the prev / next take arrows stay on the frame, and tapping the frame opens it full size.
Edit a scene — ⋯ → Edit scene opens the scene sheet (a right drawer on desk, a bottom sheet on a phone) with its title, text and Pose; save closes it, and Write and queue again on the card then writes that scene's still from the new text (the earlier still stays as a take).
Poses in Story — each scene's pose comes from its words. With a language model connected, the writer may also name a pose or describe the limbs; the app keeps a named pose only when the scene's words support it, never draws a posture the words contradict, and composes a new skeleton from the limb description when no named pose fits. Each of the four offered cards shows a small figure of the pose its still would be drawn in. The scene sheet's Pose (⋯ → Pose…) shows what will be drawn, and a pose you pick or draw there always wins.
A man as the lead — Story's own lines and built-in scenes are worded for him when the Cast record or the bible's look says he is a man.
Part comes from the Cast. A Cast with no Part is written from its own name, look and notes — there's no default archetype, and a previous Cast's Part doesn't carry over.
Beats show queue position and progress (#2 in queue, Rendering · 45%, Animating · 30%). Retry N flagged redoes every failed still and every pose or face miss in one tap. Restart story asks first — stills and clips stay in the Gallery.
Reel — Cut film is the primary action (cut options folded under Cut options), Download beside it; the player is compact and beats sit three to a row.
Phase strip: Queue → Animate → Cut. After stills, an Animate → Cut coach nudges clips before Cut film.
Lightbox: beat-to-beat prev/next + requeue chrome (same pattern as Day).
Outfit for stills — one row (kit or your photo, and the shoes) with Choose… opening the shared Clothing sheet (desk + phone); the pick rides as Image 2, and photo stills also get an Image 3 pose wireframe from the beat text (intimate + social layouts). Story has no check switches — its pose and face checks always run — so Quality is the Engine's Good / Best.
Pose variety — the scene writer is told which poses the last beats used and asked for four different ones; if its options still repeat a pose (each other, or the last beat), it is asked once more with the clash named, and the more varied set wins. The writer can name the exact layout (pose: { layout: "cook" }). Each scene sheet's Pose preview shows the guide the next queue or retry will draw, with Change pose and Try another. Each take keeps its own pose and face scores; if a new take is clearly worse than an earlier one, the card switches back to the better take (never away from one you tapped).
Continuity — each still remembers what it showed (outfit, place, light, from the scene writer's own description), and the next still keeps the same outfit, hairstyle, location and lighting unless the beat clearly changes them (a new place, a wardrobe change, a time jump, undressing).
On adult ratings, literary euphemisms in still prompts are rewritten to direct anatomy before queue (clean stories are left as written); Settings Render realism defaults to photoreal so wireframes don’t turn stills drawn.
Fal extend-video when parent is on Fal CDN; else last-frame I2V.
Tone and content rating controls (see features — Story). Adult heat requires the NSFW generator env flag.

6. Close the loop¶

Cut film in Day or Story records firstFilmCut metrics.
Dashboard Save film to Cast CTA opens Day when the cut is not stamped yet (then Watch on Cast).
Watch film on Cast, then Cut another Day film for the habit loop.

Adult content safeguards¶

Adult stills and clips (Day Suggestive / Intimate / Raunchy, Story Suggestive / Sultry / Explicit / Raunchy, and anything else whose prompt is sexual or nude) only ever show adults. Two safeguards apply; neither can be switched off.

Age wording, always. Every person on an adult still is given an explicit, mature age in the prompt itself — one short sentence such as "Both are adults — the woman in her late twenties, the man in his forties — with mature adult faces and bodies." It uses the Cast's Age trait when one is set (an Early 20s Cast is still said as late twenties; nothing younger is ever written) and in their thirties for partners, companions and anyone without one. On the compact Rapid recipes it sits right after the pose sentence, on the long briefs as the second line; WAN / LTX clips made from an adult still repeat the still's own sentence. Youth-coded words are taken out of adult text before it is queued — from beats, the Story bible and your own words: girl / boy become woman / man, ages under 25 and early twenties become the late twenties, petite, tiny, young, baby face, teen, school and classroom themes are neutralised (ordinary words like girlfriend or old-school are left alone). Rapid AIO and the Lightning engines run at CFG 1, where the negative prompt has no effect, so the age has to be in the positive wording; on engines that do read a negative (CFG > 1), child, minor, teen, teenager, childlike, young-looking, schoolgirl, schoolboy, petite youthful body, baby face are added to it as well.
Adult check on the finished still. When an adult Day or Story still lands, the vision model is asked one strict-JSON question — do all people shown clearly look like adults over 21? — before the still is shown anywhere. While it is asked the card says Checking… and no image (nor any live preview) is shown. A no, an unsure answer or a low-confidence yes withholds the still: it is hidden from the card and marked withheld in the Gallery, so it never appears in the Gallery, films, exports or sync (the server keeps no copy). The slot or scene is then queued once more with a stronger age sentence; if that take is withheld too, nothing more is queued and the card says "Withheld: the picture did not read as clearly adult — try a different seed or beat". The check is independent of Auto-review and the other Play-check switches. It uses the vision model from Settings → LLM (or LLM_VISION_MODEL, or one found on the LLM server); Settings → Play checks shows whether it is active. With no vision model configured, adult stills are shown unchecked — the age wording still applies.
Suggestive stays clothed. Suggestive is a clothed mood (lingerie, loungewear, a robe — flirty, never nude). Every Suggestive still says so in the positive wording — "Her clothes stay on, covering her chest and hips." after the clothes sentence — because Rapid runs at CFG 1, where a negative prompt does nothing; and no Suggestive beat says a robe is loosely tied or falling open, or names bare thighs or cleavage and skin (a robe always has something named under it). The adult check on a Suggestive still also asks, in the same question, whether anyone's chest, genitals or buttocks are bare; a yes is treated like a pose miss — the still is withheld, never shown as Suggestive, and the slot is queued once more with a mandatory coverage line at the top of the prompt and the beat's bare-skin words said clothed (a robe loosely tied becomes belted at the waist over a camisole, falling open over lingerie becomes over a bra and panties, bare thighs go) — on the replays the line alone changed nothing, the words did. A second bare take stops with "Withheld: the picture showed bare skin a Suggestive still must not — try a different seed or beat". Underwear, lingerie, a swimsuit or a robe over underwear count as covered; cleavage, legs, a bare back or midriff do not trigger it.

On 111 clothed adult Day stills the question raised one false alarm (under 1%).

Dashboard metrics¶

The Play film loop card on /dashboard shows:

Metric	Meaning
Film start → first cut	Days from first Film start to first Cut
Cut rate / Save-to-Cast rate	Local observability funnel
Funnel step chips	Deep-links to resume Look, Outfit, Day, etc.
Stall banner	Where you are stuck before first cut + CTA to that step
Films per week	Cuts in the last 4 weeks ÷ 4 (shown once a cut is recorded)
Stills passed review	Share of auto-reviewed stills that passed first time, with kept / requeued / flagged counts
Slowest phase	Average time per visit in Look / Outfit / Day — the phase worth optimizing. A visit over 2h is treated as "walked away" and ignored

Empty state: Open Film + Heal & ready link.

Action	How
Resume	Film resume state + `lookPackId` restore on Cast; Continue on Play
Share link	Copy share link embeds pack in `/play#lookpack=…` (large packs → Export JSON)
Cross-machine	Studio backup JSON or look pack export/import
Character mismatch	Switch to that character or restart at Look

Durable keys: play-campaign-v1 (Film resume), comfy-play-metrics-v1, play-series-v1 (Seasons), look packs on Cast + session moodboard-look-pack-v1.

Play chrome & habits¶

Chrome	Where	Behavior
Habit nudge	Dashboard, desk Film hub, Mobile Studio (not remounted on `/m/film`)	~24h after latest cut → “Tomorrow’s Day” / same-look remix; dismiss stores `comfy-play-habit-nudge-v1`
Persistence triad	Film hub (compact), Look after extract (compact, desk + phone)	This session’s look · saved on Cast · resume step
Soft-advance	Look → Outfit (with Day alternative) / Day / Story, Outfit → Day	3s cancellable countdown — Go now / Go to Day instead / Stay here

| Celebrate · manual Watch | Day / Story after first Cut | Celebrate owns Watch / Save — no auto soft-advance to Cast | | Cast home readiness | Character home + roster | Status strip (plate · looks · films); No plate chips; plate upload soft-advances to Outfit | | Play handoff glue | Look/Outfit → Day; Day/Story chrome | Adult boards realign on handoff; Day/Story status strips; Capture→Outfit soft-advance | | Engine banner | Film tools when engine ≠ ComfyUI | Warns that film stills queue on Comfy; Heal & ready + Engine settings |

Desk Film dock keeps Film · Look · Outfit · Day (+ Story after first cut). Cast, Gallery, Queue, Settings, Profile, and All tools live under More.

Mobile vs desk¶

Surface	Role
`/m` (Mobile Studio)	First-class film loop: Cast → Look → Outfit → Day → Story
`/m/film`	Phone Film hub — same resume / starter / sample reel as desk `/play`, with mobile path remaps
`/m/moodboard` · `/m/fitting` · `/m/day` · `/m/story`	Touch-first Look / Outfit / Day / Story — stills + clips, Cut film, Save to Cast
Desk	Optional large-screen handoff (Film stepper, full Story chrome)

Phone is a first-class film loop with the same vocabulary as desk. Queue and Gallery sit under More on the phone dock.

Play film guide¶

When to use Play¶

Step-by-step¶

1. Open Film (/play · /m/film)¶

2. Look (/moodboard)¶

3. Outfit (/fitting)¶

4. Day (/day)¶

5. Story (/story)¶