Moosky AI field guide
LTX 2.3 Prompt Playbook: Worked Patterns for Cinematic, Product, and Animated Video
The LTX 2.3 prompt playbook: five worked patterns for cinematic, product, and animation shots, with honest credit costs, on the live LTX model.

On this page
On this page
You found the recipe library. This is the playbook that goes with it: five working prompt patterns for the LTX 2 family, each one shown as a finished prompt, annotated clause by clause, and priced in credits before you spend a single one. People search for it as the LTX 2.3 prompt guide, and the honest naming note up front: Moosky hosts the current generation of the line, LTX-2.5, so every pattern below renders on the live model rather than a retired checkpoint. The prompting method carries across the whole family.
The distinction from the pages you may already have bookmarked matters, so here it is plainly. The LTX 2 prompt guide is the copy-paste recipe library: short patterns, grab one and go. The general AI video prompt generator guide covers prompt anatomy for every model at once. This piece owns the LTX-specific layer in between: the notation the model reads, the per-use-case structure (cinematic, product, animation), what each pattern costs at list price, and how to iterate without torching your balance.
Quick answers
What makes a good LTX 2.3 prompt? Describe change over time, not a scene. Name what moves, how the camera behaves, and what is heard, spread across the clip's full duration. LTX honors plain-language direction better than keyword lists.
How does LTX handle dialogue? You write the spoken line inside double quotes and describe the delivery in the same sentence. Speech, ambience, and effects render in one pass, lip-synced, with no separate audio step.
What is the LTX audio notation? Double quotes for lip-synced speech, angle brackets for on-screen sound effects, parentheses for background music. Weave all three into the running prose instead of bolting on an "Audio:" section.
How much does an LTX clip cost? Credits only: 100 per dollar, no subscription. A 4-second 720P draft runs about 66 credits; an 8-second 1080P hero clip about 170. The composer quotes the exact figure before you press generate, and failed runs refund themselves.
Is LTX good for animation? For animating a still with controlled motion, yes. For expressive cartoon action, it is the wrong tool in the family, and the last pattern below says exactly when to reach past it.
Read the notation before you write the sentence
LTX-2.5 renders picture and sound together, and it parses three pieces of notation to know which words are which:
- " " around a line of speech makes it lip-synced dialogue. Keep the exact words you want heard, and describe the mouth, jaw, and tone in the same beat.
- <> around a sound ties it to on-screen action:
<>boots grind on wet gravel<>is a Foley cue, not a caption. - () around music asks for a score under the shot. Leave it out when you want ambience only.
Two more rules decide most bad takes. Order your clauses the way a director gives direction: subject action first, camera behavior second, sound cues third. And size the prompt to the clip, roughly 60 to 140 words for a single continuous shot, stretching toward 180 only when dialogue and sound need the room. A prompt that describes a photograph gets a photograph with a wobble in it. A prompt that spends words on the seconds between the first frame and the last gets a shot.
That is the whole dialect. The patterns below apply it.
Pattern 1: The cinematic dialogue take
The job LTX was built for: a character delivering a line, lip-synced, with room tone, in a single pass. Structure it as a three-beat clip: who and where, the line and its delivery, the camera's one move.
A weathered fisherman sits mending a net on a dock at blue hour, his breath faintly visible in the cold air. He looks up toward the camera and says, "Storm's coming in from the north, best tie your boat down." His jaw works the consonants slowly, voice low and gravelly, eyes staying on the lens. The camera holds a slow push-in from medium shot to close-up over the line. <>Ropes creak against wooden pilings and water laps the hulls<> behind his voice.
Why it works: the line is quoted exactly, the delivery description sits in the same beat so the lip-sync has something to follow, the camera gets one verb instead of three, and the bracketed ambience is diegetic to what is on screen. Budget it at 8 to 10 seconds, 1080P: roughly 170 to 210 credits, per the composer's own quote. Draft the motion at 720P first (about 121 credits for 8 seconds), then re-render the keeper at final resolution.
Pattern 2: The atmosphere establishing shot
No character, no line, one mood, and a camera that behaves. Establishing shots fail when the prompt stops at scenery; the fix is to schedule the movement across the full duration.
Fog rolls low across an empty baseball diamond at dawn, floodlights still buzzing on. The camera tracks slowly left along the foul line, infield dirt glistening, bleachers drifting into view through the mist. A single plastic bag tumbles across the warning track in a light wind. <>Distant traffic hum and the electric buzz of the floodlights, no music, no voices<>. The fog thickens as the clip ends, swallowing the bases.
Why it works: every sentence changes something on screen, the track is the only camera verb, and the sound clause explicitly rules out a score, which matters because the notation treats parentheses as a music request and silence as the default you should ask for. This is the cheapest keeper on the board: 8 seconds at 720P lands around 121 credits, and most atmospheric shots survive their draft pass at that price.
Pattern 3: The product-ad hero shot
Product clips are where re-rolls burn real money: the label must stay legible, the light must stay flattering, and the motion must stay small enough not to warp the object. Anchor a start frame and keep the camera to one move.
[Attach a clean product photo as the start frame] The glass serum bottle on a wet slate surface catches a moving highlight as the camera performs a very slow push-in, droplets sliding down the condensation. The label stays sharp and readable throughout. The bottle remains perfectly still; only the light and droplets move. <>A soft glass chime as the highlight passes, and faint studio room tone<> underneath. Background stays deep charcoal, shallow depth of field.
Why it works: the start frame locks identity, the sentence "the bottle remains perfectly still" is doing more work than any adjective about premium feel, and the single chime cue gives the edit a sync point. Six seconds at 1080P runs about 130 credits. For a three-shot ad, render all three at 720P draft cost first (three takes under 400 credits total), then final-res only the winning angle.
Pattern 4: The animated start-frame transition
The animation query people bring LTX is usually "animate this still," and LTX answers it well as long as the motion is controlled: a portrait turning, smoke curling, hair lifting in wind. First-to-last-frame control is the trick: give the clip an opening and a closing image and prompt the transition between them, keeping identity, palette, and props stable.
[Attach a stylized illustration as the start frame and its dusk color-variant as the last frame] The storybook fox on the hilltop blinks, then turns its head toward the setting sun as the light warms from gold to amber. Grass bends progressively left in rising wind. The illustration style, line weight, and colors stay identical; only the head turn, blink, and grass movement change. <>A high windrise and a soft bell toll from the distant village<>.
Why it works: naming what must not change protects the style the way naming what moves protects the motion, and the FFLF pair removes the model's temptation to invent. Be honest about the ceiling: LTX is built for talking scenes and controlled motion, not big athletic action or squash-and-stretch cartoon performance. If the brief is genuinely expressive animation, run the comparison in the LTX vs Seedance vs HappyHorse guide first; several family siblings do characterful motion better. For still-to-life with restraint, this pattern is the economical one: 8 seconds at 720P, about 121 credits per attempt.
Pattern 5: The multi-shot continuity sequence
LTX was built to keep a subject and a setting stable across cuts, which makes one run of several shots possible without the character drifting between them. Label the beats and give every beat its own framing, action, and camera verb:
Shot 1: A barista in a green apron steams milk at the counter, steam billowing past her forearms, camera static at hip level. <>The steam wand shrieks briefly<> over low café chatter. Cut to Shot 2: same barista, same café, now a close-up as she wipes the steam wand, glancing up at something off-screen, camera a slow rack focus from her hands to her face. Cut to Shot 3: wide of the quiet morning café, the barista in background soft focus, camera a slow pull-back toward the door. <>Espresso machine ticks, a door chime rings at the end<>.
Why it works: the repeated anchor ("same barista, same café") is what multi-shot continuity reads on, each beat is a complete instruction rather than a vibe, and the chime lands on the final second like an edit point. Fifteen seconds at 1080P prices around 310 credits, so storyboard first: talk the beats through the ChatGPT drafting loop for free, and reverse-engineer any reference clip you are imitating with the video-to-prompt workflow.
The honest credit math
Moosky bills LTX by the clip on a 100-credits-per-dollar balance, and the composer shows the exact number before you run. These are the list prices of the patterns above, so you can budget a session instead of guessing:
| Clip | Resolution | About | Use |
|---|---|---|---|
| 4 s | 720P | 66 credits (~$0.66) | idea smoke test |
| 8 s | 720P | 121 credits (~$1.21) | motion draft for patterns 2 and 4 |
| 6 s | 1080P | 130 credits (~$1.30) | product hero (pattern 3) |
| 8 s | 1080P | 170 credits (~$1.70) | dialogue draft (pattern 1) |
| 10 s | 1440P | 303 credits (~$3.03) | final, high-pixel |
| 10 s | 4K | 472 credits (~$4.72) | delivery master |
The shape of a cheap session follows from the table: draft at 720P, lock the take, then spend the 1440P or 4K credits once, on the shot you kept. Those top tiers cap clips at 10 seconds; 1080P runs to 20. Nothing is charged for a failed render, it refunds itself, and unspent credits do not expire. If the quote on the run button bothers you, the LTX pricing page walks the buying model end to end.
Running the playbook
Pick the pattern closest to your shot, swap your subject, setting, and line into its skeleton, and keep the clause order intact. The recipe library has the short-form versions for quick grabs, and the LTX-2.5 composer is where every prompt above renders, cost quoted first.
Frequently asked questions
Is LTX 2.3 free? The weights are open to download, but running them takes a workstation-class GPU, setup, and electricity. Hosted access is where the money question actually lands: on Moosky you pay per clip on a credit balance (100 credits per dollar) with no subscription, the exact price shows before every run, and failed renders refund.
Where do I paste an LTX 2.3 prompt? On the LTX-2.5 lane of the video generator. It is the current LTX generation Moosky hosts, and it runs everything in this playbook including the bracketed audio notation.
How long should an LTX prompt be? One paragraph, typically 60 to 140 words, up to about 180 when spoken dialogue and layered sound need room. Length is the wrong knob; coverage of the clip's duration is the right one.
Why did my LTX clip come out with no sound? Check that audio generation is enabled on the run, then check your notation: quoted lines lip-sync, angle brackets mark on-screen effects, parentheses request music. A sound described outside those markers reads as scenery, not as a cue.
Why did my character change between shots? Multi-shot prompts need explicit anchors. Repeat the identity clause ("same barista, same café") in every beat, as Pattern 5 does, and keep each beat's framing, action, and camera verb separate so the model is not forced to guess what continuity means.
Can these patterns run on other models? The three-block structure travels well, but the audio notation is LTX dialect. For dialogue with a longer single shot, the Seedance-family guide shows which model to swap into the same sentence.
More from Moosky AI
Create with Moosky AI
Turn ideas into finished productions
Access leading AI models and production workflows from one workspace.
Try Moosky AI →Comments
Comments are reviewed before they appear publicly.
Join the discussion —sign inorcreate an accountto leave a comment.
No approved comments yet. Start the conversation.