User Manual

Every owntent feature in one place. For each one you'll see what it does and where to find it. New here? Start with the steps below.

Getting started

1Create a new project from a logline or synopsis.
2Shape the story in the AI thread and Develop tabs, building out scenes and characters.
3Generate and refine shots in Storyboard, Control, and the Pipeline.
4Finish the timeline in Preview and Edit, then export your video.

Supported models · providers

Image generation

god-tibo-imagen (Codex / GPT Image)

Keyless

Default image path. Reuses local Codex/ChatGPT auth (~/.codex/auth.json), so no API key is needed. Standard path for keyframes, concept art, and sheets.

Route · location: /api/owntent/imagen

OpenAI GPT Image (gpt-image-2.0 / gpt-image-1)

BYOK

Direct OpenAI image generation (BYOK). Kept only as the legacy character-reference path; new screens use /api/owntent/imagen.

Route · location: /api/owntent/image

Higgsfield image (Nano Banana Pro/2, Soul, Cinema Studio)

Keyless

Runs the local Higgsfield CLI server-side. Relies on host CLI login, not a BYOK key. Used for face/identity consistency (soul-id), cinematic stills, and reference work.

Route · location: /api/owntent/higgsfield

fal.ai image (FLUX.2 Pro, FLUX 1.1 Pro, Seedream v4, Nano Banana 2)

BYOK

Calls many image models via fal.ai's single queue API (BYOK fal key). FLUX family, Seedream, Nano Banana for general/high-quality/reference editing.

Route · location: /api/owntent/fal

Google Gemini image (Nano Banana Pro / 2)

BYOK

Direct Google Gemini image models (BYOK google key). Up to 3 reference images for image-to-image. Pro = high quality, Flash = fast/cheap.

Route · location: /api/owntent/gemini

xAI Grok Imagine (Quality / Standard)

BYOK

Direct xAI Grok Imagine image models (BYOK xai key). Quality = 2k high quality, Standard = fast standard quality.

Route · location: /api/owntent/grok

Atlas Cloud NSFW image (Flux Dev NSFW, Seedream v5 NSFW)

BYOK

Uncensored (NSFW) image generation (BYOK atlas key). Shown only when NSFW mode is on; safety checker disabled.

Route · location: /api/owntent/atlas

ComfyUI (self-hosted SD/Flux)

Keyless

Queues workflows to the user's self-hosted ComfyUI. BYO — no models bundled; any SD/Flux model/workflow the user chooses. Always shown regardless of NSFW toggle.

Route · location: /api/owntent/comfyui

Custom BYO model (OpenAI-compatible / REST)

BYOK

Calls an arbitrary user-registered image-generation endpoint (BYOK). Also the extension path for providers without a built-in integration (e.g. Suno).

Route · location: /api/owntent/custom-model

Video generation

PoYo — Seedance 2.0 / Fast / Sora 2 / Sora 2 Pro

BYOK

Video generation via PoYo's queue API (BYOK poyo key). Seedance 2.0 (cheapest 720p I2V), Seedance 2.0 Fast ($0.08/s), and OpenAI Sora 2 / Sora 2 Pro (audio, T2V/I2V).

Route · location: /api/owntent/poyo

Higgsfield video (Cinema Studio 3.5 / 3.0, Veo 3.1)

Keyless

Video via the Higgsfield CLI (keyless, needs host CLI login). Cinema Studio = top-tier cinematic, Veo 3.1 = photoreal + audio.

Route · location: /api/owntent/higgsfield

fal.ai video (Kling 3.0/2.6, Seedance 2.0, Hailuo 02)

BYOK

Many video models via fal.ai's queue API (BYOK fal key), including Kling cinematic I2V, ByteDance Seedance 2.0, and MiniMax Hailuo.

Route · location: /api/owntent/fal

Atlas Cloud NSFW video (Wan 2.2/2.5/2.6, Kling v3, Vidu Q3)

BYOK

Uncensored (NSFW) video generation (BYOK atlas key). NSFW-mode only. Wan Spicy family, 1080p Kling/Vidu, etc.

Route · location: /api/owntent/atlas

Seedance 2.5 (ByteDance)

Keyless

Coming soon. To be enabled once the official API spec is confirmed; not selectable yet.

Route · location: catalog only (coming soon)

Text / LLM (script, prompt authoring, vision)

OpenAI / Anthropic Claude / Google Gemini (LLM + vision)

BYOK

BYOK LLM text/vision generation (openai/anthropic/google). Used for script, shotlist, prompt authoring, and reference-frame analysis (vision). A 'custom' provider allows any OpenAI-compatible endpoint.

Route · location: /api/owntent/llm + /llm/stream + /vision

Local CLI agents (Codex / Claude Code)

Keyless

Uses a locally installed Codex / Claude Code CLI as the text-generation agent (no key). The keyless alternative on the LLM path. Vision not supported.

Route · location: /api/owntent/llm (agent param)

Voice / narration (TTS)

ElevenLabs TTS (Multilingual v2, Turbo v2.5, v3 alpha)

BYOK

BYOK ElevenLabs text-to-speech (narration, dialogue dubbing). Exposes standard params (language code, stability, style, speed). 29 languages, emotional delivery.

Route · location: /api/owntent/elevenlabs/tts

Music / BGM

ElevenLabs Music

BYOK

BYOK ElevenLabs Music for per-scene BGM/soundtrack (text-to-music). The primary BGM provider since the Suno public API is unreleased.

Route · location: /api/owntent/elevenlabs/music

Suno (v4)

BYOK

A Suno key slot and catalog entry exist, but there is no dedicated server route (public API unreleased). Users add it via the BYO model form and call it through /api/owntent/custom-model.

Route · location: BYO model form

Other — model/asset sources

CivitAI (model browse/download)

Keyless

CivitAI model browse/download proxy. Search/metadata is keyless; only some gated downloads need a token. For sourcing checkpoints/LoRAs to use in ComfyUI, etc.

Route · location: /api/owntent/civitai

Runway / Kling (direct)

BYOK

BYOK key slots for runway and kling are defined, but there is no dedicated direct route; Kling models are currently served via fal.ai (/api/owntent/fal).

Route · location: via fal.ai (/api/owntent/fal)

BYOK note

BYOK key slots are the 12 defined in keys/registry.ts (PROVIDER_KEYS): openai (llm,image), anthropic (llm), google/Gemini (llm,image,video), fal/fal.ai (image,video,audio), higgsfield (video — but the actual integration is a keyless CLI), poyo/PoYo Seedance 2.0 (video), runway (video), kling (video), elevenlabs (audio), suno (audio), atlas/Atlas Cloud NSFW (image,video), xai/xAI Grok (image). All keys are stored locally only (localStorage 'owntent:api-keys' / key vault) and are never sent to or stored on our server (local-first BYOK). Keyless paths: god-tibo-imagen (/api/owntent/imagen, reuses local Codex auth), Higgsfield CLI, local Codex/Claude Code CLI, ComfyUI (user-run), and CivitAI browse.

Foundations · auto-editing pipeline

Under the hood, generation never goes straight to a timeline. A set of deterministic, LLM-free guards run first, then a base timeline is assembled and trimmed, and finally optional AI edit decisions refine the cut. These quality patterns are re-implemented in pure TypeScript inside the owntent-core package; each one cites the open-source project it draws from.

Foundations · applied patterns

Several of the quality and assembly patterns are adapted from the open-source executive-producer project OpenMontage and re-implemented as pure, LLM-free TypeScript so they run instantly with zero credits.

Executive-producer stage contract

Every pipeline stage (brief, script, shotlist, keyframes, clips, assemble, render, and more) carries a declarative rubric: what it produces, the review focus a person should check, and concrete pass criteria. It also flags which stages should pause for human approval, turning a loose flow into a reviewable, gated contract.

Slideshow-risk gate

Before any paid asset is generated, the plan is scored across six dimensions — repetition, decorative visuals, weak motion, weak shot intent, over-reliance on typography, and unsupported cinematic claims. The overall verdict (strong / acceptable / revise / fail) catches output that would feel like a slideshow instead of directed video, and surfaces on the Evaluate tab as Structure Quality.

Delivery promise

The project locks what it promised to deliver — motion-led, stills-led, source-led, or text-led — and then checks that the shot plan actually honors it. If a motion video silently drops below the required ratio of real-clip shots, it is flagged as a quiet slideshow downgrade so you can stop before compose instead of shipping a weaker cut.

Variation / structure check

A deterministic structural pass scans for the concrete patterns that produce generic, repetitive output: lazy filler language (English and Korean), empty or duplicated scene descriptions, one location dominating the plan, and the same framing or camera reused across shots. It reports a repetition score that feeds the slideshow gate, all without any model call.

Cost tracking

Paid steps follow an estimate → reserve → reconcile ledger: each operation gets a pre-estimate, reserves budget before running, and settles to the real cost afterward. A budget mode (off / warn / cap) decides whether to allow, warn, or block the next step, keeping BYOK spend visible and governed.

Real-clip / stock search

Beyond generation, the pipeline can search free stock-footage providers (Pexels, Pixabay) for real motion clips and normalize each result to a common shape — download URL, thumbnail, dimensions, duration, and attribution. This opens a genuine real-video path that the auto-assembler can drop straight into the timeline.

Auto-editing pipeline

From generated assets to a finished cut, the auto-editing pipeline runs as an ordered sequence. Each step is grounded in real code — quality gates first, then assembly, trimming, transitions, optional AI edit decisions, and finally preview and export.

1
Generation
A resumable, ten-stage pipeline takes a project from brief through script, dialogue, shot list, style bible, keyframes, clips, and audio. Stages run only when the medium needs them — still or prose media skip the motion stages — and gated stages wait for human approval before spending image or video credits.
2
Consistency & quality gates
A deterministic consistency pass (no LLM) detects scene-mappable problems — empty descriptions, lines spoken by someone not in the cast, missing portraits, knowledge-graph issues — and returns exactly which scenes to regenerate. Alongside it, the slideshow-risk and delivery-promise gates block low-quality or quietly downgraded plans before compose.
3
Auto-assemble base timeline
The best take or keyframe for each shot is laid end-to-end in narrative order to build a base timeline in one pass. Clips that fail to insert do not advance the cursor, so no empty gaps are left, and take URLs are resolved only through a trusted lookup so a model can never inject a forged source.
4
Smart / auto trim
Optional smart trim samples each clip at low resolution and removes leading and trailing black, fades, and frozen (static) sections, always keeping a minimum length and falling back to the full clip if analysis fails. An opt-in frame-analysis mode can instead use a vision agent to choose the best in/out points per clip.
5
Transitions
Edit points between cuts are planned automatically, choosing a transition technique, duration, and direction for each boundary based on whether it stays within a scene or crosses into a new one. The same engine drives both heuristic and AI assembly, and each point can be previewed, retimed, or overridden, then re-applied to the timeline.
6
AI edit decisions (EDL)
When an AI backend is available, the timeline context is sent to the model, which returns an Edit Decision List — which take to keep per shot, trim in/out, and durations. The result is always repaired and validated against the real candidates and never throws; if the model output is missing or invalid, it falls back to a deterministic heuristic edit.
7
Preview & export
The assembled cut plays back in the embedded video editor, where you can scrub edit points, tweak transitions, and re-assemble. When the cut is final, the project is rendered and exported to a finished video file.

All features

Per-feature detailed guides. Open an item to see overview, how-to, models, and tips.

Explore · Main

How it works

Location: /how-it-works

Pricing

Location: /pricing

Account

Location: /account

Settings

Location: /settings

Bring your own GPU (0 credits)

Location: /settings

Projects

Location: /projects

Production

Autopilot

Location: /studio/[project]/pipeline

Control

Location: /studio/[project]/control

Storyboard expands each scene as an accordion, managing every shot as a thumbnail + detail alongside the scene summary, characters, and dialogue. Each shot shows its keyframe or generated clip (hover preview + a versions modal when present), with reorder (up/down + drag), edit, and generate buttons. In edit mode you set cinematography fields — shot size, angle, lens, lighting, camera motion, transitions — via selectors, and can apply an auto cinematography suggestion derived from the scene's emotional arc and beat.

Location: /studio/[project]/storyboard

Canvas

Location: /studio/[project]/canvas

Getting started

Supported models · providers

Image generation

god-tibo-imagen (Codex / GPT Image)

OpenAI GPT Image (gpt-image-2.0 / gpt-image-1)

Higgsfield image (Nano Banana Pro/2, Soul, Cinema Studio)

fal.ai image (FLUX.2 Pro, FLUX 1.1 Pro, Seedream v4, Nano Banana 2)

Google Gemini image (Nano Banana Pro / 2)

xAI Grok Imagine (Quality / Standard)

Atlas Cloud NSFW image (Flux Dev NSFW, Seedream v5 NSFW)

ComfyUI (self-hosted SD/Flux)

Custom BYO model (OpenAI-compatible / REST)

Video generation

PoYo — Seedance 2.0 / Fast / Sora 2 / Sora 2 Pro

Higgsfield video (Cinema Studio 3.5 / 3.0, Veo 3.1)

fal.ai video (Kling 3.0/2.6, Seedance 2.0, Hailuo 02)

Atlas Cloud NSFW video (Wan 2.2/2.5/2.6, Kling v3, Vidu Q3)

Seedance 2.5 (ByteDance)

Text / LLM (script, prompt authoring, vision)

OpenAI / Anthropic Claude / Google Gemini (LLM + vision)

Local CLI agents (Codex / Claude Code)

Voice / narration (TTS)

ElevenLabs TTS (Multilingual v2, Turbo v2.5, v3 alpha)

Music / BGM

ElevenLabs Music

Suno (v4)

Other — model/asset sources

CivitAI (model browse/download)

Runway / Kling (direct)

BYOK note

Foundations · auto-editing pipeline

Foundations · applied patterns

Executive-producer stage contract

Slideshow-risk gate

Delivery promise

Variation / structure check

Cost tracking

Real-clip / stock search

Auto-editing pipeline

Generation

Consistency & quality gates

Auto-assemble base timeline

Smart / auto trim

Transitions

AI edit decisions (EDL)

Preview & export

All features

Explore · Main

How it works

Pricing

Account

Settings

Bring your own GPU (0 credits)

Projects

Production

Autopilot

Control

Canvas

Blocking (3D)

Voice & BGM

Tools

Story

AI Threads

Develop

Import Script

Ontology Packs

World

Dialogue

Webtoon

Read

Factions

Chapters

Ontology

Graph

State

Foreshadowing

Graph Query

Screenplay Analysis

Evaluate

Skill optimizer