UGC pipeline adds autopilot content flow with a side-by-side manual editor

• bas_fijneman's latest demo shows the UGC tool running an automated A/G flow while keeping a side editor open for manual changes.
• Earlier posts from the expanding pipeline thread and the full autopilot update show the system settling into a repeatable path: generate assets, assemble clips, draft captions, then queue publishing.
• The screenshots in the editor screenshot thread and the pipeline dashboard screenshot suggest the builder is treating this less like a single video toy and more like a small production stack, with Influencers, Pipeline, Media, Schedule, and Results views.
• bas_fijneman said in a reply about prompting that good outputs still depend on very specific prompting, and a reply about model inconsistency adds that video generation can still fail even with detailed prompts.
• Distribution is still the unresolved part: a publishing reply says posts are pushed as drafts and published manually, while a waitlist source reply says early signups were coming from Facebook and WhatsApp groups.

You can see the interface evolve from an earlier editor build to the newer autopilot demo, with the pipeline view exposing image, video, assemble, publish, and track as separate stages. There is also a live waitlist signal in the signup source reply and the 96-signup post, which ties the content machine back to an actual app, PufferPages.

The clearest change across the posts is structure. By the earlier pipeline thread, the tool already had a timeline, manual caption editing, and a fully automated mode. By the later demo, bas_fijneman says setup was run manually for control, while A/G was fully automated.

The pipeline screenshot attached to the pipeline dashboard screenshot breaks the flow into five stages:

• Image
• Video
• Assemble
• Publish
• Track

That staging matters because the full autopilot update claims the system can run fully autopilot or manually, not just generate isolated clips.

The editor looks built for intervention, not just review. The UI shown in the editor screenshot thread includes fields for hook ideas, hook text on the reaction clip, text on the base video, caption writing, audio settings, and a sequence timeline.

The same screenshot also shows two synchronized preview panes, Reaction and Base Video, plus caption blocks on the timeline. That matches the latest demo, where bas_fijneman describes the editor on the side as full control.

The builder is blunt about the weak point. In a reply about prompting, bas_fijneman says the workflow is doable, but you need to know how to prompt the videos, and testing costs a bit. In a follow-up on inconsistency, he adds that current generators still have a level of inconsistency and can mess up even when the prompt is very specific.

One concrete tool choice does surface in a model reply, where he names Seedance 2.0 as part of the stack.

Autopilot stops short of true hands-off distribution. The main autopilot post says the system should produce content for TikTok and Instagram, but a later reply says the accounts are older, the tool pushes posts as drafts, and publishing is still done manually.

There is a second operational wrinkle in an account warm-up reply, where bas_fijneman says he used standard warm-up tactics and still needs a better way to handle US accounts. another reply says he wants to build an auto warm-up system, but is not sure it will ever work.

The last useful reveal is that this content engine is already feeding a live audience test. In the 96-signup post, bas_fijneman says a new app hit 96 waitlist signups within a week, before launch. a reply on source channels says those signups were coming from Facebook and WhatsApp groups, not X.

That detail makes the whole demo more concrete. The pipeline is being built around PufferPages, which is named in the editor screenshots from the editor interface, and the content is being used to find distribution before the product is even live.