The whole studio, feature by feature
Everything it takes to make a show worth coming back to.
Showspring is a brand-agnostic AI production studio for episodic shows, built for quality and consistency, not volume. Every feature here serves one goal: a high-quality show your audience returns to, made by a crew of agents you direct. You take an idea from premise to a finished, publishable episode in one workspace, and a persistent show bible keeps characters, voices, and storyline consistent from one episode to the next. You point it at your own show and your own channels, not a platform your audience logs into. The Doodle Cast is one show built with it.
◆ Quality, not volume
Built to make a show worth watching.
The internet does not need more disposable AI. Showspring is built so the output holds up across a whole series, with you directing every step. Every safeguard here protects quality, it does not crank volume.
On-model, every time
Characters, locations, and props are pinned to reference images and a no-new-objects rule, so nothing drifts or morphs mid-scene.
A produced look, not a prompt look
Scenes are composed shot by shot, and a recurring segment locks its own visual direction, so episodes feel made, not merely generated.
It learns what holds attention
Retention from your own published episodes feeds the next idea, so quality compounds across the season instead of guessing.
01 One tool that makes whole episodes, not clips
The episode pipeline
You take an idea from premise to a finished, publishable episode in one workspace. A show bible that builds from what you make keeps every character in voice from one episode to the next.
Multi-model idea pitching, judged and yours to override
Give the Creative Director step a direction and it proposes four distinct premise themes, then has two writers draft a full pitch from each. A judging panel scores the pitches and names what is strongest about each. You pick the winner, and the approved premise seeds the narrative spine every later beat follows.
Bible-aware scripts that keep every character in voice
Scripts read the show bible before they write: personalities, speech styles, recurring segments, and format rules. Tag characters as mandatory cast and the writer must give each real dialogue on a minimum number of clips, or the script is rejected and rewritten. Choose the writing model per episode.
Per-character voices and an instant readout pass
Each character carries an ElevenLabs voice plus a stored speech style. Before you commit to a full voice render, a readout step previews the script with browser text-to-speech so you catch pacing and awkward lines for free. Characters marked silent work through visuals alone.
Multi-engine visuals with consistent characters and props
Scenes start as still images from your chosen engine, then animate into clips. A locations and props database with reference images keeps recurring sets and objects on-model, and each clip can be assigned its own locations and props.
Script version history with compare and restore
Every script generation is snapshotted. Browse the history, open a side-by-side compare of any two versions, activate an older one, or restore from trash. Each snapshot keeps its feedback notes so you can see why a version changed.
Edit-timeline export for DaVinci Resolve
When you want to finish in a real editor, Showspring exports an OpenTimelineIO bundle that DaVinci Resolve, Premiere, and other editors import. Do the AI assembly in Showspring, then polish in the NLE you already use.
02 Concept to scheduled vertical video in one batch
The shorts and vertical video generator
A concept becomes finished 9:16 clips in one self-contained batch: idea, script, still image, animation, audio mix, burned-in captions, and scheduling. Every clip is shot vertical for phones, and the workflow is built around how short-form actually gets watched.
Batch idea and script generation grounded in your channel
Spin up 5 to 20 shorts at once. Idea generation reads your show bible, your channel research, and your channel's own retention insights, plus a strategy you pick: subscriber growth, viral reach, brand awareness, engagement, storytelling, or let the system choose. The same strategy threads through idea, script, and per-clip regeneration.
Four locked script fields per clip
Each clip's script splits into four enforced fields. A scene description is a still image only, with no motion verbs. A character description is a static pose. A video-action field is the timed motion beat sheet, with per-second timestamps. Dialogue is one spoken line, kept short. Regenerate any single field or clip without re-running the batch.
Two stages: still image, then timed animation
Image and video are separate, sequential steps. The scene and character descriptions produce the 9:16 starting frame; that frame plus the beat sheet drive the animation. A no-new-objects rule can hold the shot on-model, so nothing materializes mid-clip.
Character voice, multi-track mixing, and burned-in captions
Dialogue is voiced in the character's own voice, then mixed under the video with a music bed. Each clip has independent volume, mute, and trim controls for the video audio, the character voice, and the music. Captions are burned into the render so the short reads with the sound off.
Built for the loop
Shorts replay automatically and replays count toward watch time, so the script step is told to design motion such that the final frame rhymes with the first, same camera and a similar pose, so the loop feels intentional rather than a hard cut.
Recurring segment templates with locked direction
Tie a batch to a recurring segment and every clip inherits its creative direction: a start-image look, a motion structure, and audio direction. A recurring format keeps a consistent opening frame, beat structure, and edit style across episodes.
Timezone-aware scheduling from your own performance data
After rendering, a recommend-schedule step buckets your top performers by local day and hour and proposes a date and time per clip with plain reasoning, spreading clips across days and rotating through your best hours. Adjust any slot or accept the plan.
03 Audio-first shows on the same bible
The podcast workflow
A dedicated audio-first pipeline for podcast-style episodes, from idea through scripted dialogue, per-character voices, intro and outro music, and a final mixed audio file. It reuses the same characters and show bible as the video pipeline, so a podcast sounds like the same cast.
Several ways to start an episode
Pick a mode: let the system recommend a topic, bring your own idea, just make one, reuse a saved idea, or run a debate. You set a target length in minutes and the script aims to fill it.
Debate mode with real research and a strict-facts toggle
Debate mode does live web research first, builds a dossier, and writes a debate-ready concept anchored to it. A strict-facts toggle tells the cast they may hold strong opinions and joke, but must not invent statistics, quotes, dates, names, or studies that are not in the research.
Per-character voices and music beds
Dialogue is voiced per character through ElevenLabs, and the episode tracks intro and outro music so a finished show has its own open and close. Everything assembles into a single final audio file.
Narrative arc, not a list of talking points
Even a debate episode is required to have an arc: a starting position, cracks that open, a real confrontation, a shift, and a landing. A pitch that just lists things the hosts will argue about gets rejected in favor of one with a through-line.
04 Make, publish, then make better
Analytics and the feedback loop
Showspring pulls performance data back from your published content so the next episode is informed by the last one. A tabbed dashboard shows what worked, and that same retention data feeds the idea and script steps.
Tabbed dashboard with deltas and sparklines
Overview, Shorts, and Episodes tabs show per-platform metrics over a selectable time range. Delta chips compare the current window against the prior one and suppress noise on low counts, and sparklines show the trend at a glance.
Retention curves per clip
Watch-percentage curves show where viewers drop off. When enough curves exist, Showspring rolls them into channel-wide insights about which structures hold attention and which lose it.
A publish-time heatmap
Performance is bucketed by day of week and hour in your local timezone so you can see which schedules tend to land. It is the same signal the shorts scheduler uses to propose slots.
Retention feeds the next idea
The idea and script steps read the retention insights as direct feedback: replicate the structures of strong-retention shorts and avoid the patterns of weak ones. That closes the loop from making to publishing to analyzing and back to making.
05 Bring trusted reviewers into the build
Review and collaboration with Director Copilot
Director Copilot lets you invite trusted reviewers to follow an episode build live and leave feedback, while keeping them strictly read-only. Reviewers watch at a separate viewer site; you see everything in an inbox inside the editor.
Read-only by design
Collaborators reach only the viewer site and a small set of allowed actions. A guard enforces the boundary: they can follow the build and comment, but they cannot edit the script, regenerate, or otherwise change the episode. Reviewers never reach the creator app.
Comments anchored to the work
Feedback attaches to specific targets: a clip, a generated image, a video take, or a script line. Reviewers can add a reaction to a comment, and you can resolve comments as they are handled.
Two-way reply threads
Each comment supports a reply thread so you and the reviewer can go back and forth. Display adapts to who is reading: you see your own replies as Director, the reviewer sees theirs as themselves, and new replies appear live.
Voting on takes and a live inbox
Reviewers can vote on competing video takes, and you see comments, unseen counts, presence, and vote tallies grouped by clip in an inbox, with jump-to-shot to land on the flagged item. Presence and new feedback update on a short poll, so it feels live without video calls.
+ The rest of the studio
Everything else under one roof
The pieces that make the five workflows above hold together, episode after episode.
Workspace and access
- Multi-project workspaceSwitch between multiple shows in one workspace. Each project has its own bible, characters, and accent-color theming, and queries are scoped per project.
- Per-project access controlAdmin-configurable access restricts accounts to specific projects. Sandbox accounts cannot create new projects.
- Cost visibility and sandbox policyAdmin-only usage views break spend down by model and callsite over a time range. A sandbox policy restricts which models cheaper accounts may use.
Characters, locations, and bible
- Character profiles with voicesEach character stores a name, role, personality, visual, speech style, a voice, a voice summary, and reference portraits. Changing a voice refreshes the preview.
- Locations and props databaseFull create, edit, and delete for locations and props, with multi-image uploads and sortable galleries. Clips can be assigned multiple locations and props.
- Show bible with synced video knowledgeA per-project bible plus a video-knowledge index auto-synced from your YouTube uploads, with character mentions tracked. The bible builds from what the show produces rather than being authored before episode one. Optional in-universe post format per project.
Generation engines
- Selectable script modelsWrite scripts with Gemini, Grok, Claude, or a local Llama model, chosen per episode. Sandbox accounts default to a single cheaper model.
- Multiple image enginesGenerate stills with local SDXL, Qwen, Gemini, ZTurbo, Google Flow, or the Higgsfield bridge. Sandbox accounts are limited to Gemini and Google Flow.
- Video with job pollingVideo generation submits a job and polls for status until the clip is ready. The Higgsfield bridge runs on a local GPU and is gated off for sandbox accounts.
Publishing and distribution
- YouTube publishing and native schedulingShorts and episodes publish to YouTube through the API, with immediate publish or native scheduling, and AI-content disclosure when you enable it.
- Cross-platform metadataPer-platform titles, descriptions, captions, and hashtags are generated for YouTube, X, Facebook, Instagram, and TikTok. X chains off the YouTube publish; TikTok, Instagram, and Facebook go out through a Social Champ bulk CSV export.
- Scheduling queue with proposalsA shorts queue tracks queued and proposed states with proposed dates and per-platform timing, recorded per clip and per platform.
See it on a real show.
Showspring runs The Doodle Cast end to end, episode after episode. Take a look, then tell us about your show.