Podcast Design Canvas

Repository: e35dev/podcast-design-canvas-3

Create a self-serve visual podcast production workspace where creators transform synced multi-speaker raw recordings into polished, personalized, publishable video episodes without needing a traditional editor.

This repository is maintained against the product direction below. The maintainers use this document, VISION.md, and CONTRIBUTING.md as the standard for accepting or closing work.

Who It Serves

• Podcast creators who record with Riverside-style separate synced speaker tracks
• Solo hosts who want professional results without learning a complex editing suite
• Podcast teams and agencies producing repeatable show formats for multiple episodes or clients
• Power users who want to design
• save
• reuse
• and eventually monetize custom podcast layouts

Product Workflows

• Create a new episode by importing a Riverside link or uploading separate synced video files for each speaker
• then assign each file to clear speaker buckets such as Host
• Guest 1
• and Guest 2.
• Add host and guest social links during setup so the product can understand names
• topics
• references
• brands
• and likely transcript spellings before generating the edit.
• Choose a preset visual style with layout and pacing options
• preview how the episode will look
• and apply it without needing to manually position every element.
• Open a canvas editor to build or customize a reusable podcast layout by dragging and layering speaker video frames
• shapes
• backgrounds
• captions
• title elements
• b-roll areas
• and overlays.
• Clean and balance episode audio with simple controls for noise reduction
• leveling
• enhancement
• and speech clarity
• presented as creator-facing quality choices rather than technical audio settings.
• Use contextual editing tools to add captions
• b-roll overlays
• visual callouts
• title moments
• and short-form-style engagement patterns at key moments across a full-length episode.
• Save a finished layout or style as a reusable show template so future episodes can keep the same identity while still adapting to each episode's speakers and topics.
• Export a polished long-form video episode that feels deliberately edited
• visually coherent
• accurately captioned
• and ready to publish.

Intended End State

A creator can go from raw synced podcast tracks and a few social links to a finished, professional-looking long-form episode with clean audio, accurate text, personalized context, engaging visual moments, and a reusable visual identity for the show.

Product Taste

• The product should feel like Canva adapted to podcast production: visual
• direct manipulation
• simple defaults
• and creative freedom for advanced users.
• The default experience should emphasize preset quality: users should be able to get a polished result by choosing from clear style
• layout
• and pacing options.
• The pro experience should expose flexible canvas controls for custom layouts
• layered shapes
• branded frames
• captions
• overlays
• and reusable templates.
• The system should support many podcast identities rather than a single house look: every show should be able to feel distinct.
• Visual edits should feel professional and intentional: clean framing
• coherent layouts
• readable captions
• tasteful overlays
• and rhythm that keeps a long episode engaging.
• Social context should make the edit smarter: better transcript spellings
• relevant b-roll choices
• more accurate references
• better on-screen context
• and captions or titles that fit the people speaking.

Accept Work That

• Merge clean PRs that pass CI, match the Vision Model, and improve an accepted workflow or quality bar.
• Prefer small coherent changes that can ship immediately over broad speculative rewrites.
• Treat product taste and user workflow fit as first-class acceptance criteria.
• Summarize merged work as product progress, not as raw PR activity.

Close Work That

• Close PRs that are incomplete, off-vision, overlapping, stale, or likely to create product drift.
• Do not leave requested-change queues by default. Close with clear resubmission guidance.
• Close technically correct PRs when they solve the wrong problem or move the product away from the captured vision.
• If a PR is promising but messy, explain the clean smaller PR that should be submitted next.

Riverside-Style Link Import

Episode setup can start from a Riverside-style episode link instead of manual uploads: paste a share link that references the synced speaker tracks (directly, or via an episode manifest) into the "Import from a Riverside-style episode link" field and click Import. Host, Guest 1, and Guest 2 populate with playable media that flows through the same preview and export pipeline as uploads.

The repo's declared sample Riverside-style manifest link (also shown next to the import field in the app) is:

https://riverside.fm/studio/pdc-sample-episode

Its episode id resolves to the bundled sample episode in app/riverside-sample.js — three real synced VP8/opus WebM speaker tracks (animated speaker cards with motion, name tags, running timecode, and distinct audio tones), embedded as self-contained data:video/webm URLs. Resolving the sample therefore performs no file or network reads at all: it imports identically over http(s), file:// with local-file access, and plain file:// (where Chromium CORS-blocks local fetches), with a clean console. Regenerate the sample with node scripts/make-riverside-fixtures.mjs. Links may also reference an episode manifest .json path or name tracks directly. Real Riverside network integration and account sign-in are intentionally deferred: the importer never fetches remote media — only app-relative paths, in-session blob: object URLs, and self-contained data: video URLs are importable, and the wrapping share URL is parsed as text, never requested.

Scoring Labels

This repo does not require Gittensor scoring labels on merged PRs. The registry uses default_label_multiplier: 1.0, so accepted merged PRs score without a gittensor:* label.

Maintainer-only operational labels such as active-step labels may exist, but contributors should not add labels to make work score.

Verification

• Run or preserve typecheck before submitting product work.
• Run or preserve lint before submitting product work.
• Run or preserve test before submitting product work.
• Run or preserve preview-build before submitting product work.