AI Tools that Power
our Video Production

Four internal AI tools, two production workflows — how R&D powers TheSoul Publishing's content engine.

0
Internal Tools
0
Workflows
0%
Hit Prediction
0× Faster
vs Manual AI Workflow
TL;DR
In 30 seconds.
01
Production core: Faber
Multi-agent super-tool. Auto-finds viral content on YouTube, runs the full production cycle, and publishes Shorts and Long-form. Long-form can be semantically cut back into Shorts and republished. Works human-controlled or ~90% autonomous.
02
Hit Replication
Two engines reproduce viral content. Faber (full production from YouTube source) and SlopMachine (faster, reference-driven shortcut).
03
Original Content
For in-house scripts and concepts. Faber for fully-produced cinematic output, SlopFlow as the faster node-based alternative. Ersa scores scripts before they enter production.
04
Status
Faber 2.0 shipped · SlopFlow + SlopMachine in active use · Ersa MVP scoring scripts · Ersa's next step: feedback loop into Faber for smarter virality detection.

Two pipelines, one engine.

Different starting points, four engines, one closed loop. Faber's published videos feed back into Ersa, which then retrains Faber.

Pipeline 01
Hit Replication
Find & adapt viral content
Via Faber · full production
YouTube Shorts
auto-discovered by Faber
Faber
Manual post-editing
human polish
Publish Short
Via SlopMachine · reference-driven shortcut
Hit video reference
SlopMachine
Manual post-editing
Publish Short
Pipeline 02
Original Content
In-house scripts & concepts
Via Faber · cinematic production
Idea / Script
↑ hit probability
Ersa
Faber
Manual post-editing
Branch by output format
If Short
direct publish
Publish Short
If Long
~5 min episode
Publish Long
Faber
semantic cut
Publish Shorts
Via SlopFlow · node-based fast alternative
Idea / Script
SlopFlow
Manual post-editing
Publish
Closed loop
Faber ↔ Ersa
Feedback loop

Ersa only analyses Faber's videos. Each Faber output's real-world performance feeds back into Ersa — which then tells Faber which YouTube ideas to mine and which script patterns produce hits.

Faber
publishes a video
↻ measure performance
Ersa
learns from results
↑ smarter ideas & scripts
Faber
next iteration

Explore the system map.

Drag any node, scroll to zoom, click filters to isolate a workflow. Four internal tools, two pipelines, and the Ersa → Faber feedback loop.

Drag · pan · scroll to zoom

Four tools.

A multi-agent super-tool, two node-based shortcuts, and a hit scorer.

Faber

Ready · v2.0
Multi-agent super-tool · Both pipelines

The R&D production core. Auto-discovers viral YouTube content, runs the full agentic production cycle, publishes both Shorts and Long-form, then can cut a Long semantically back into Shorts. Built by R&D.

10hper 5-min Long
3×vs manual AI
4-5×night speedup
~90%autonomous
What it does
  • ▸Auto-finds viral videos on YouTube
  • ▸Runs full agentic production cycle: script → scenes → render
  • ▸Publishes Shorts and Long-form natively
  • ▸Semantically cuts Long-form back into Shorts
Two modes
  • ▸Human-controlled — tight quality control, every step approvable
  • ▸Autonomous — ~90% hands-off run
Faber V2 Studio dashboard

SlopFlow

Ready
Original content · Node-based

Simpler, faster node-based alternative to Faber for original content. Visual workflow editor — build your generation pipeline by connecting nodes. Built by Dmitrii Petrov.

node-basedvisual editor
fastlight pipeline
  • ▸Drag-and-drop pipeline construction
  • ▸Fast iterations for original concepts
  • ▸Not used for hit replication
  • ▸Authored by D. Petrov
SlopFlow node-based editor

SlopMachine

Ready
Hit replication · Reference-driven

Simpler, faster alternative to Faber for hit replication. Reproduces viral content by working directly from a video reference. Built by Dmitrii Petrov.

batchcloning at scale
fastlow-touch
  • ▸Drop a reference URL — get a remade video
  • ▸High throughput for viral catch-up
  • ▸Authored by D. Petrov
SlopMachine processing queue

Ersa

MVP
Hit prediction · Feedback loop

Increases the probability of producing a hit. Scores scripts and concepts before they enter Faber production. Future direction: closed feedback loop into Faber for smarter virality detection. Built by R&D.

64%prediction
68parameters
1382training channels
Today
  • ▸Scores hook, script, visuals, channel history, trends
  • ▸Improves the script before Faber generation
Next
  • ▸Feedback loop into Faber — picks more viral YouTube concepts
  • ▸YouTube engagement data flows back into script quality
Ersa dashboard with training stats

Tool × Workflow.

Tool Hit Replication Original Content Note
Faber
Multi-agent super-tool · full production · publishes Shorts + Long · cuts Long → Shorts.
SlopMachine
Reference-driven hit replication shortcut. Built by D. Petrov.
SlopFlow
Node-based original content alternative. Built by D. Petrov.
Ersa
Increases hit probability. Hit already validated in replication.