OneVideo Studio Guide
From one idea to a finished film
16 sections covering both creation modes, how to write a premise that actually shoots, how to pick a model, what everything costs, and what to do when a film goes wrong.
Before you start
OneVideo turns one sentence, or a finished script, into a vertical short film with characters, dialogue, subtitles and music. You don't need editing skills or source footage.
New accounts get 400 credits
That's enough for one complete 15-second film. See section 10 for how credits are spent.
| Capability | Free | Member |
|---|---|---|
| Maximum length | 15s | 15 / 30 / 45 / 60s |
| Quality | Model standard tier | Standard or member HD tier; actual resolution follows the selected model |
| Concurrent projects | 1 | 3 |
| Character portraits | Placeholder (no visual anchoring) | AI-generated, locks the character's face |
| Watermark | Yes | No |
| Queue priority | Standard | Priority |
About character portraits
Portraits are what keeps a character's face consistent across shots. The free tier uses placeholders that don't participate in visual anchoring, so faces may drift between shots in multi-shot films. A film with one character throughout needs membership.
Two creation modes
This is the most important choice before shooting. It determines the film's structure, its cost, and how much you can intervene.
| One-click | Director mode | |
|---|---|---|
| How it shoots | One continuous take, minimal cuts | Shot by shot, then assembled |
| Visual continuity | Better — no seams inside one take | More seams, stitched via anchors and tail-frame chaining |
| Control | Low, the system decides | High — inspect, edit and reshoot any single shot |
| Credits | Charged at target length | About 1.45× one-click |
| Best for | First film, fast output, single-mood content | Multi-scene drama, brand films, shot-level control |
Why director mode costs more
To fit all the dialogue, director mode usually overruns your target length by about 45%, and you're charged for real seconds. One-click is hard-capped at 15s per segment so it never overruns.
- 1Start with one-click for your first film. Move to director mode once you know what you want.
- 2Director mode earns its cost when you need to reshoot one shot without redoing the film.
Your first film
The fastest reliable path, about sixty seconds of setup.
Open Create
The plus button in the bottom bar, or the create entry at the top.
Write one sentence
One person, one place, one thing happening. No strategy, no backstory.
Keep the defaults
15 seconds, vertical 9:16, one-click, standard resolution. Leave advanced options closed.
Check the estimate
The Action panel shows estimated cost and your balance. If it's short, you're told exactly how much before anything is created or charged.
Shoot, then leave
A 15-second film usually takes 4–8 minutes. You can close the page and come back.
Won't shoot
"Discuss the financing difficulties facing small businesses" — no person, no place, no action. The model has nothing to render.
Will shoot
"Late at night in a shut-down electronics factory, the owner sits alone by a silent machine, thumb hovering over a loan app, unable to press it."
Write a filmable premise
The video model reads literally. It won't infer what you mean — it renders what you wrote. The more visible and specific, the more accurate the result.
- 1Write what is visible. "He's anxious" can't be filmed; "he keeps checking his phone, thumb hovering without pressing" can.
- 2One action per sentence. Six things in one line get compressed into a few seconds of visual mush.
- 3Avoid metaphors and optical descriptions. "Tears burst the dam" may render an actual dam. The system rewrites some of these automatically, but writing it right is more reliable.
What safety review rejects
Every model has its own content policy, and they differ a lot. The default model refuses intimate or suggestive imagery outright — the shot fails, though a rejected shot is never billed. If your premise is adult, pick the model marked Overseas at creation time; it is built for it. What no model will ever generate: anything sexual involving minors, non-consensual acts, or real public figures. Those are refused at the writing stage, before any credits are spent.
Creative brief (optional)
Before shooting you can expand your sentence into an editable set of characters, settings and conflict. It calls a model and costs 5 credits, but it stops the AI from guessing wrong and wasting a whole film.
Getting the model to actually shoot it
Every rule below comes from frame-by-frame comparison of real output — what the script asked for versus what the model delivered. These are not style preferences; they are hard execution limits.
| Limit | Ceiling | What happens past it |
|---|---|---|
| Distinct actions per shot | ~2 per 10 seconds | The third one usually gets dropped — and it is normally the last beat you wrote, i.e. your payoff |
| Dialogue line length | Under 8 words | Longer lines get paraphrased, shortened, or skipped |
| Single shot duration | 15 seconds | Hard ceiling on every link; longer stories must be split across shots |
- 1Put the turn in the first two thirds of the shot, never on the final beat. Whatever you write last is what gets dropped first.
- 2Use ordinary names. In testing, the invented name DRAY came out as "Dre" on one model and "Drake" on another; Serena and similar common names never broke.
- 3Never write "cut to", a flashback, or a time jump inside one shot. A single generation is one continuous take — the model cannot edit, and the shot gets flagged as unshootable.
- 4Lock key props on three axes: material, location, exclusivity. "Handprints on the door" can render as blood-red prints down the whole corridor; "frost-white, on the inner face of the door only, nowhere else" renders correctly.
- 5Make characters who share a frame visually unmistakable — different hair, different clothing colour, different build, a distinct marker each. The closer they look, the more likely the model merges them or swaps one mid-shot.
Renders as nothing
"Their breathing quickened", "clothes half undone", "fingertips traced the scar" — literary phrasing puts nothing on screen. The model does not read subtext; it renders only what is literally, physically described, and when the description is vague it picks the most conservative option: two people standing still, close together.
Renders
"Her shirt is pulled open to the third button, left shoulder bare", "his right hand closes over her wrist and pins it to the glass" — say whose hand, which part of the other person it touches, and how it moves.
Actions these models handle badly
Zero-gravity float (comes out as ordinary walking), multi-step precise timing (a door closing while someone passes through and turns back), and mirror-reflection consistency (the reflection rarely matches the live action). Avoid these, or give each its own shot.
These rules matter more with your own script
With "use my own script" the AI does not rewrite anything — write it vaguely and it gets shot vaguely. In testing, a 25-character custom script left the model free to improvise across a full 15 seconds, and the result matched almost nothing the author intended.
Choosing a model
The models differ far more than their spec sheets suggest, and they differ in places you would not expect. Below is what the same script produced on each link.
| Need | Use | Why |
|---|---|---|
| Multi-shot continuous story | The workhorse model — 15s per shot, 480p / 720p tiers only | The only model that genuinely continues from the previous shot — measured, the next shot's first frame is near pixel-identical to the previous tail frame |
| Single shot, best image quality | The model whose quality tiers include 1080p | The only 1080p you can buy anywhere on the platform, and excellent single-shot quality — but it cannot continue from a previous shot |
| Adult content | The model marked Overseas in the picker | A dedicated link; the others refuse this material |
| Talking-head avatar | Switches automatically when you enable Avatar mode | Requires an avatar and voice to be configured first |
The 1080p model cannot do multi-shot narrative
Measured directly: it accepts the previous shot's tail frame and then ignores it completely. Every shot effectively cold-starts, so faces, wardrobe and lighting are left to chance — the same character can come back as a different person in the next shot. Use the standard-tier workhorse model for continuity; save the 1080p one for one-take pieces.
Continuity can still break occasionally
Even with the right model, the previous shot's frame is sometimes rejected by the model's own safety filter as a "real person", and that shot falls back to text-only generation without pixel continuity. When this happens the project page says so explicitly — it never fails silently.
- 1Higher tiers cost more, and the delivered file size follows the tier you bought: 480p / 720p / 1080p produce 854×480, 1280×720 and 1920×1080 respectively (swapped for vertical).
- 2A shot refused by content review is not billed for video generation, but the script and storyboard steps already spent are not refunded.
- 3Switching models does not preserve shots you already generated. To compare models, start a new project rather than retrying inside the existing one.
Choose a genre
A genre isn't just a label. Each one carries its own screenwriting logic — hook formula, pacing, taboos, ending style. Picking the right one makes a visible difference to script quality. Leave it unset and the system infers one from your premise.
| Genre | When to use it |
|---|---|
| Sweet Romance | Campus, workplace, mutual crushes — driven by emotional tension |
| Emotional | Family, parting, regret — the ending should linger |
| Suspense | A reversal in the last 3 seconds; everything before is setup |
| Sci-Fi | Large-scale imagery, non-human entities, space or apocalypse |
| Underdog | Someone dismissed proves them wrong — fast, dense payoffs |
| Comedy | Absurd premises and unexpected turns |
| Horror | Atmosphere and restraint, not gore |
| Healing | Slow pace, small details, quiet kindness |
| Commercial | Dramatising a product or service benefit |
| Fashion Runway | Clothing, catwalk, texture-led imagery |
| Rebirth | A second chance, rewriting fate |
| Family Drama | In-laws, couples, generational conflict |
| Period Drama | Court intrigue, status, indirect dialogue |
| Viral Skit | Short, sharp, strong hook — built for feeds |
| Business Advisory | B2B lead-gen content, four-act structure, compliance-constrained |
Business Advisory genre
Purpose-built for lending and financing advisory content: a fixed four-act structure (problem → warning → method → outcome), two recurring characters, advisor rather than broker positioning, plus built-in compliance checks that block guaranteed-return claims, artificial urgency and naming specific financial institutions.
Use your own script
Already have a script? Switch to "Use my script" and the dialogue is preserved word for word, with no AI rewriting.
- 1Shot lists — headings like "Shot 1 | Wide" with visual description, camera and sound notes
- 2Standard screenplay — scene headings (INT. OFFICE - NIGHT) with character names on their own line
- 3Table shot lists — "1 | Close-up | 3s | finger presses the button | VO: countdown begins"
- 4Prose — no shot markers; dialogue detected by quotation marks, shots split by paragraph
- 5Plain voiceover — one continuous block, split evenly by punctuation
What never becomes dialogue
Technical annotations (camera, lighting, sound, transitions, duration) only affect the image — they're never spoken and never burned into subtitles. Title cards are skipped entirely. This is what stops a heading like "## Visual bible" from being read aloud.
Durations are recalculated
The seconds you write are treated as a hint. The system redistributes time based on each shot's actual dialogue load, so every line finishes naturally and silent shots don't waste screen time.
Settings reference
Everything you can adjust before shooting.
| Setting | Options | How to choose |
|---|---|---|
| Length | 15 / 30 / 45 / 60s | 15–30s for feeds; 45–60s for full narrative. Free tier is 15s only |
| Aspect ratio | 9:16 / 16:9 / 1:1 | 9:16 for TikTok/Douyin/RED; 16:9 for YouTube landscape |
| Quality tier | Model-specific | The card shows the actual output (480P, 720P, 768P, 1080P or 2K) and its rate. HD is members-only |
| Subtitles | Classic / TikTok / Karaoke / Pill / Off | Classic white is the safest default. Bilingual available |
| Music | On / Off | Native-audio drama mixes low-volume mood music under the voices |
| Reference image | Upload one | The film starts from this frame; AI keeps the character and style consistent |
| Shot-by-shot approval | On / Off | Pauses after each shot for your confirmation. See section 12 |
| Batch variants | 2+ films | Same premise, multiple versions for A/B testing. Charged per variant |
Getting reference images accepted
Upload a character portrait or establishing shot and the system starts from it. With photos of real people, note that the video model has a privacy policy on recognisable faces — a tight close-up is likely to be rejected, while a medium or wide shot passes far more often.
Credits and pricing
The rule is simple: calling a large model costs credits, system functions don't. Re-editing is free, regenerating costs, re-rendering is free.
| Quality | Rate | Availability |
|---|---|---|
| Standard (480P–768P) | 8–14 credits / second | All users |
| HD (720P–2K) | 14–28 credits / second | Members only |
| Length | One-click standard | One-click HD | Director standard | Director HD |
|---|---|---|---|---|
| 15 秒 | 120 | 210 | 176 | 308 |
| 30 秒 | 240 | 420 | 352 | 616 |
| 45 秒 | 360 | 630 | 528 | 924 |
| 60 秒 | 480 | 840 | 696 | 1218 |
| Item | Credits | When it happens |
|---|---|---|
| Script writing | 15 | AI writes the script. Not charged when you supply your own |
| Shot breakdown | 10 | Turning the script into a shot list |
| Character portrait | 12 each | Members only. Not charged if generation fails |
| Shot first frame | 4 each | Locks the starting image of a shot |
| Creative brief | 5 | Only if you expand your premise |
Free of charge
Re-rendering the film, changing subtitle style, editing script text, automatic retries when QA fails, and any image that falls back to a placeholder — none of these cost credits.
If your balance is short
You're stopped before shooting, with the exact shortfall and a top-up link. No project is created and nothing is charged — your settings are preserved. You can also drop to a 15-second version instead.
What happens during production
After you hit shoot, the project page tracks progress live. The pipeline runs in this order.
Script
Story structure and dialogue, following the genre's logic. Skipped if you supplied a script.
Shot breakdown
Split into shots. Consecutive beats in one location merge into a single take to reduce seams.
Pre-flight
Checks whether the dialogue fits, whether shots are filmable, and compliance.
Character portraits
One portrait per character, the single source of truth for their appearance.
Shooting
The longest stage. Shots generate in parallel, roughly 3–5 minutes each.
QA
Frames are sampled and compared against the portraits and the script. Failures are reshot automatically at no extra cost.
Assembly
Concatenation, audio mixing, subtitle burn-in and loudness normalisation. Local, no model calls.
Delivery
Cover image and share page generated. Ready to download or publish.
| Status | Meaning | What to do |
|---|---|---|
| Queued | Waiting for a free generation slot | Wait; you can leave the page |
| Generating | Writing, shooting or rendering | Wait |
| Awaiting confirmation | Paused in shot-by-shot approval mode | Confirm or edit that shot |
| Needs review | QA flagged something | Read the QA note, reshoot or accept |
| Partially failed | Some shots didn't render | Click retry failed shots |
| Completed | Film is ready | Download, share or publish |
| Failed | The run stopped | Check the reason; most are retryable |
How long it takes
About 4–8 minutes for 15 seconds, 15–25 minutes for 60 seconds. There may be a queue at peak times. You can close the page after shooting starts and come back when the film is ready.
QA and redoing work
Every shot is checked after it renders: does the character match their portrait, does the image match the script. Scores start at 100 and deduct; below 70 is flagged for review. The score measures whether something was shot wrong, not artistic merit.
| Action | What it does | Credits |
|---|---|---|
| Edit | Change this shot's description or dialogue text | Free |
| Regenerate | Reshoot this shot with the current description | Charged by shot length |
| Re-render | Reassemble the film without reshooting | Free |
- 1The usual pattern is edit the description first, then regenerate. Regenerating without changing anything mostly returns a similar result.
Re-rendering has a deadline
Re-rendering depends on the original shot footage, which stays available for 24 hours. After that, re-rendering fails and the film must be regenerated from scratch. Handle subtitle and music changes the same day.
Export and publish
Four actions are available once a film completes.
| Action | What it does |
|---|---|
| Download MP4 | Downloads the file. No watermark for members |
| Share | Opens the system share sheet — straight to messaging apps or the camera roll |
| Save as template | Stores this film's setup for reuse on similar content |
| Publish | Posts to the Explore page for other users to see |
Public playback page
Every finished film gets a public playback link you can send to a client or colleague. They don't need an account to watch.
Advanced features
Worth knowing once you're producing regularly.
- 1Shot-by-shot approval — pauses after each shot so a bad one never propagates through the rest of the film. Costs your attention, saves credits on important work.
- 2Batch variants — shoot several versions of one premise for A/B testing before you spend on distribution.
- 3Sequels — build on a finished project and inherit the full cast, wardrobe and voices, so a series stays consistent.
- 4Templates — save a setup that worked; swap the premise and reuse everything else. The biggest time saver for daily publishing.
- 5Asset library — every portrait, first frame and QA frame is kept, downloadable, and reusable as a reference image.
Troubleshooting
What you see, why it happens, and what to do about it.
| What you see | Why | What to do |
|---|---|---|
| Insufficient credits | Balance won't cover this film | Top up, or shorten the film / drop to 480p |
| Shot refused | Triggered the video model's safety policy | Rewrite that shot; remove intimate, violent or suggestive elements |
| Character's face changes | No portrait anchoring (free tier), or that shot rendered independently | Upgrade, or use one-click to reduce seams |
| Garbled Chinese on screen | Image models can't render accurate Chinese characters | Avoid close-ups that rest on text; blur it or leave it unwritten |
| Dialogue cut off | The line is longer than the shot's duration | Shorten the line, or increase the length setting |
| Long queue | Peak platform load | Wait; you can close the page. Members get priority |
| Re-render fails | Source footage is past its 24-hour window | Regenerate the film from scratch |
| Some shots failed | Individual shots timed out or were refused | Click retry failed shots — only those are reshot |
Known limits
Things that genuinely can't be done today. Stated plainly so you don't spend time trying.
- 1Chinese text inside the image. Phone screens, documents and signage render as characters that only look like Chinese. This is a hard limitation of image models.
- 2Close-up reference images of real faces. The video model has a privacy policy on recognisable faces and will reject them regardless of where the image came from.
- 3Scene detail across shots. Faces hold, but furniture and props in the same room may change model between shots. Mood and lighting stay consistent.
- 4Exact lip sync. Dialogue and mouth movement are approximate, not frame-aligned.
- 515 seconds per shot. Longer films are assembled from multiple shots, and seams show small changes.
The safest way to work
Test with 15 seconds, 480p, one-click first. Once the direction is right, shoot the full version. That costs 120 credits to find out — far less than 696 for a 60-second director-mode film.
Ready to shoot?
Test with 15 seconds at 480p first — 120 credits to find out whether the direction is right.
Create nowStill stuck?
Message us on Telegram — a human reads every message.