Generation Guide
The Generation Engine is the heart of Lomio — it orchestrates AI model calls to produce scripts, storyboards, image prompts, and more, with full task tracking and credit metering.
Overview of the generation system
Every AI generation in Lomio goes through the same pipeline regardless of output type:
- Task creation — you configure a generation task (type, inputs, parameters) and submit it.
- Queue — the task enters the Workflow Engine queue. Generation is asynchronous; you do not wait for it in real time.
- Execution — Lomio's AI Runtime selects the appropriate model, constructs the prompt using the Prompt Engine, and calls the model via the APIPod gateway.
- Post-processing — the output is parsed, validated against the output schema, and stored.
- Credit deduction — credits are deducted from the workspace balance and a usage log entry is created.
- Result delivery — the generation panel updates and you receive a notification (if enabled).
Generation is asynchronous
You can close the browser or navigate away after submitting a task — it continues running in the background. Results appear in the Generation tab when complete.
Generation types
Script generation
Generates a full episode script including scene headings, action lines, and dialogue. Lomio uses the project description, episode synopsis, character profiles, and knowledge-base documents as context.
To generate a script:
- Open an episode and click Generate Script.
- Select the characters who appear in this episode.
- Optionally add a director's note — a brief instruction to the AI about tone, pacing, or specific plot beats.
- Choose a script template (e.g. Standard Drama, Social Media Short Drama, K-Drama Style). Templates are prompt presets tuned for different formats.
- Click Generate. The task is queued. Typical completion time: 20–90 seconds depending on script length and model load.
Storyboard generation
Takes an approved script and generates a shot-by-shot storyboard: scene descriptions, shot types (close-up, wide, over-the-shoulder), character positions, and image generation prompts for each shot.
Storyboard generation requires an Approved or Locked script in the episode. Open the episode, go to the Storyboard tab, and click Generate Storyboard.
Image prompt generation
Generates optimized image-generation prompts for each storyboard frame. These prompts include character visual descriptors, LoRA trigger words, style tags, and negative prompts. You can use them directly with the video pipeline or export them for external image tools.
Dialogue rewrite
Rewrites a selected section of dialogue while preserving plot intent. Useful for changing tone, adapting to a different character voice, or localizing content. Select lines in the script editor and click Rewrite Selection.
Translation
Translates a script or storyboard into another language while preserving formatting, scene structure, and character names. Translation tasks appear in the Generation tab with a language tag.
Task states
| State | Meaning |
|---|---|
| Queued | Task is waiting for an available worker. Credits are not yet deducted. |
| Running | Task is actively executing. Credits are reserved. |
| Completed | Task finished successfully. Credits are deducted. Output is available. |
| Failed | Task encountered an error. Credits are not deducted for failed tasks. The task can be retried. |
| Cancelled | You cancelled the task before it completed. Credits are not deducted if cancelled before execution began. |
Reading generation results
When a generation task completes, the output appears in-line in the relevant tab (Script, Storyboard, etc.) as well as in the Generation History panel.
Each result card shows:
- The generation type and task ID.
- Date and time of completion.
- Credits consumed.
- The model and prompt version used.
- A diff view if there was a previous version — changes are highlighted.
- Action buttons: Accept (overwrite the current draft), Keep both (save as an alternate), Reject.
Credits and generation costs
Every generation task consumes credits from your workspace balance. Credit costs vary by task type, output length, and model tier. You can see the estimated credit cost for any task type on the generation panel before submitting.
If your workspace runs out of credits, queued tasks will be paused and new tasks cannot be submitted until you top up. See the Billing FAQ for how to add credits.
Retrying failed tasks
If a task fails, open the Generation History panel, find the failed task, and click Retry. The task re-enters the queue with the same parameters. Common failure causes:
- Model timeout (rare — retry usually resolves this).
- Output schema validation failure (the model produced output in an unexpected format — the system will automatically reformat and retry once).
- Content policy rejection — the prompt or inputs violated safety rules. Review your inputs and the Responsible AI Usage policy before retrying.
Cancelling tasks
Tasks in Queued state can be cancelled immediately with no credit charge. Tasks in Running state can be cancelled, but if execution has already started a partial credit charge may apply. Completed tasks cannot be cancelled — reject the output instead.
Generation history and audit trail
Every generation task is logged permanently in the workspace audit trail. This includes who triggered the task, what parameters were used, which prompt version was active, and what was produced. Workspace Owners and Admins can access the full audit log from Workspace Settings → Audit Log.