TARKOVSKY CREATIVE TOOLSOpen studio →

More possibilities.
Your decisions.

Structured shot plans

  1. Save and lock the script. Open Shot List. Keep the storyboard unlocked while adding shots.
  2. Open AI shot planner and choose a scene.
  3. Choose Tarkovsky credits or your own provider key. Enter a model ID supported by that provider. Add creative direction.
  4. For credits, estimate the maximum charge, then generate. For your own key, the provider bills your account directly.
  5. Review the descriptions, sizes, movements, and durations. Select the shots to keep, then apply them. Remaining suggestions are rejected.

Each proposal covers one scene and accepts at most 12 shots. A model may return fewer shots. Duplicate labels receive a suffix. Accepted shots append to the shot list. Existing shots are preserved. Shot plans are model suggestions, not a claim about footage already produced.

Tracked script edits

In version 0.02, tracked edits arrive inside the Story chat. Ask for a change in the chat, with Tarkovsky credits or your own key. A reply that carries edits shows a Proposed change · Scene N card: the text to delete is struck through and the replacement follows, with Accept and Reject. Accept applies the change to the working draft. Nothing is applied without a click.

The locked production script remains unchanged until you lock a new draft. On a locked draft, accepting a change needs a working draft first; select Duplicate to keep writing. A reply without edits can be appended to the working script instead.

Every replacement must match one unique, non-overlapping passage. Invalid model output is rejected. No project content changes when you only ask for a proposal. The AI shot planner above is unchanged.

AI voice synthesis

  1. Choose Tarkovsky credits, or add your OpenAI key on the Models page.
  2. Open Assets or Edit → AI voice studio.
  3. Select a built-in voice, speed, and words. OpenAI also supports delivery direction. Each request supports up to 2,500 characters.
  4. For credits, request an estimate and confirm its price. Select Generate AI voice. The request runs as a background job.
  5. Play the generated MP3. Select Add voice to audio timeline, then save the project.
  6. In Edit, set its start, trim, gain, and fades. Render the film to hear the full mix.

Tarkovsky credits fund Kokoro narration through OpenRouter or an on-demand Lambda GPU. OpenRouter is the economical default. Lambda includes startup time in its allocation price and terminates after each job. You can instead use OpenAI gpt-4o-mini-tts with your own key. The audio is AI-generated. Disclose that fact when presenting it to listeners. The feature uses built-in voices, not cloned identities. See OpenAI’s voice documentation.

The provider call is sent once. An uncertain or failed response may still be billed; check the provider console before retrying. Audio remains private project media. The final export supports up to 16 separate audio cues.

LUT import and halation

In Export, import a numeric 3D .cube file and select it under 3D color LUT. Files must be smaller than 2.5 MB and contain 2–33 samples per axis with valid sample and domain values. Unsupported, malformed, or nonnumeric files are rejected. LUTs are private to their project.

The render applies the LUT, the selected look, halation, grain, and vignette in that order. Choose a LUT designed for the color encoding of your footage; this release does not infer camera log transforms or provide a full color-managed grading pipeline. Halation adds a soft red glow around bright areas. It is an artistic approximation, not a physical film-stock simulation.

Effects appear in rendered H.264 or ProRes exports. Browser source players do not preview these effects. LUT and halation settings are stored with each render job. Choose No LUT and set halation to zero to disable them. See the FFmpeg filter documentation.

Costs and review

Shot plans and tracked edits use the same text providers and credit billing as the Story chat. Credit estimates reserve a maximum; settlement is based on the recorded provider cost. Provider calls can be charged even when the output is unusable or you reject every suggestion. No retry is automatic.

Voice offers Tarkovsky credits for Kokoro via OpenRouter, or your OpenAI key for directed speech. Select Estimate voice credits, then confirm the displayed fixed service price. Dedicated Lambda narration appears only when its account and worker are enabled. LUTs and halation run on the application renderer and do not consume model credits. Existing project, media, export-duration, and output-size limits still apply. Only project owners and editors can generate or apply these tools.