Vidfiddle 0.2.2

Vidfiddle turns your RSS feed into short vertical videos on your own server, holds each one for your approval, and schedules it. This release fixes logging in over plain HTTP.

Fixed

• You can log in from another machine on your network again. If you opened Vidfiddle over plain HTTP at any address other than localhost, such as http://192.168.1.20:3000 on a home server or http://<server-ip>:3000 on a new VPS, the password was accepted and you landed back on the login page. The Docker image always marked the session cookie Secure, and browsers throw those away on plain HTTP. The cookie is now Secure only when you reach the app over HTTPS, directly or through a proxy that sets X-Forwarded-Proto (Caddy, Traefik, nginx with that header, Cloudflare Tunnel).

Upgrading from 0.2.1

docker compose pull
docker compose up -d

No migration this time. If you pinned VIDFIDDLE_VERSION=0.2.1 in .env, change it to 0.2.2 first. In one-container mode, redeploy so the host pulls the new image.

Vidfiddle 0.2.1

Vidfiddle turns your RSS feed into short vertical videos on your own server, holds each one for your approval, and schedules it. This release fixes reposts.

Fixed

• Reposts are tracked again. Reposting a video to a platform it had already gone to got scheduled in your publisher, but Vidfiddle never saved it, so the repost didn't show up in Vidfiddle and its status was never checked. The app reported success anyway. A migration now lets reposts repeat, while each video still gets one original post per platform.
• If Vidfiddle ever can't save a repost, it now says so, and warns that trying again would post the video twice. Before, the repost dialog showed no error at all.

Upgrading from 0.2.0

docker compose pull
docker compose up -d

The worker applies the migration when it starts. If you pinned VIDFIDDLE_VERSION=0.2.0 in .env, change it to 0.2.1 first. In one-container mode, redeploy so the host pulls the new image.

Vidfiddle 0.2.0

Vidfiddle turns your RSS feed into short vertical videos on your own server, holds each one for your approval, and schedules it. This release adds two open source schedulers and a published Docker image, and drops the YouTube clips feature.

New

• Publish through Postiz or TryPost. Both are open source; run them yourself or use their cloud. They join Publer. Vidfiddle uploads each video to Postiz as a file, so with Postiz your Vidfiddle doesn't have to be reachable from the internet. Setup is under Publishing in the README.
• A published image. ghcr.io/zellerhaus/vidfiddle, for amd64 and arm64. docker compose up -d now pulls it instead of building from source, and VIDFIDDLE_VERSION in .env keeps you on a release.
• One-container mode. node scripts/start-all.mjs runs the app and the worker together, for hosts that run one service per app (Railway, PikaPods, an Unraid template). See One container.
• Bulk create. Select posts on Content and make a video from each at once.
• Running Generate again on a post reuses its video instead of failing.
• Vidfiddle no longer calls Creatify when Creatify isn't configured.

Changed

• LightPost is now experimental. Released LightPost can't post these videos yet; use Publer, Postiz or TryPost.

Removed

• YouTube clips, along with the Deepgram, Creatomate and Submagic settings. Vidfiddle now does one job: posts in, videos out. A migration drops the YouTube tables if they're empty, and renames them (youtube_sources_removed, youtube_clips_removed) if they hold anything, so no data is lost.

Upgrading from 0.1.0

git pull
docker compose pull
docker compose up -d

The worker applies the migration when it starts. You can delete WORKER_SECRET, DEEPGRAM_API_KEY, CREATOMATE_API_KEY and SUBMAGIC_API_KEY from your .env.

The eight showcase videos stay attached to v0.1.0.

Vidfiddle 0.1.0

The first public release. Point Vidfiddle at an RSS feed and it turns each post into short vertical videos on your own server, holds them for your approval, and schedules them.

What's in it

• Local video engine. OpenAI writes the scenes, draws a still per scene, reads the script and times every word; ffmpeg cuts shots with camera moves, burns in word-by-word captions and mixes music under the voice. About $0.06 for a 30-second video.
• Eight visual styles, or your own prompt: felt diorama, watercolor, 3D clay, deep space, paper-cut collage, sci-fi paperback, flat vector, photoreal.
• Pacing and music. Slow, medium or fast cuts at the same cost. A CC0 starter music pack filed by mood, your own tracks in music/, or ElevenLabs Music per video. Narration leveled to -14 LUFS with ducking under the voice.
• ElevenLabs voices as an option alongside OpenAI's.
• Auto Mode for new feed posts, with an approval queue where you approve, reject, edit the script or ask for changes.
• Captions per platform and scheduling through Publer or self-hosted LightPost.
• One docker compose up: app, worker and Postgres. Local disk or any S3-compatible bucket for media.

The showcase

The eight MP4s attached below were made end to end from NASA and NOAA posts with no editing. All eight cost $0.46 together. launch.mp4 is the launch video.

Setup is in the README. Questions go in Discussions, bugs in Issues.