← sources

We Built a Working Local AI Filmmaking Pipeline in ComfyUI — RTX 4070 12GB

source: reddit (reddit:comfyui)published 2026-10-06 18:10ingested 2026-10-07 02:42

retained text

We’re building and actively using a **working AI filmmaking pipeline**, focused primarily on **local models and ComfyUI**.

This is not just a concept or a collection of isolated workflows. We already have a production pipeline that takes a project through multiple stages:

**script → scenes → storyboard → reference-controlled images → video → voice → editing**

We’re using it to produce a real episodic project. Some of the current results can be seen here:

[https://www.youtube.com/@TikiTalkShowNet](https://www.youtube.com/@TikiTalkShowNet)

Our goal is to find out how much of the complete production process can realistically be done **locally on a single consumer workstation**, without relying heavily on closed cloud services.

**Hardware:**

NVIDIA GeForce RTX 4070 — 12 GB VRAM  
Intel Core i9-14900K  
96 GB RAM  
5 local drives

The pipeline is already working, but we’re gradually replacing and improving different stages with local tools and ComfyUI workflows.

The main challenges we’re currently working on are:

* consistent characters across many shots;
* especially **multiple consistent characters in the same frame**;
* persistent location references;
* creating coherent narrative storyboards rather than isolated good-looking images;
* storyboard → video continuity;
* reference conditioning;
* local image and video models in ComfyUI;
* LoRA / ControlNet / IPAdapter-style approaches;
* running larger models within **12 GB VRAM**;
* automatic detection and regeneration of failed shots.

At the moment, **character consistency and narrative storyboard generation are probably our biggest problems**.

We’re not posting this mainly to promote the project — we’re much more interested in meeting people working on similar local pipelines and exchanging **actual workflows, experiments, failures and solutions**.

The practical question we’re trying to answer is:

**How far can a real episodic filmmaking pipeline be pushed using local AI and ComfyUI on consumer hardware?**

If you’re working on something similar, we’d be very interested to hear what approaches are working for you — especially for **character consistency, multi-character scenes and storyboards**.

What is currently the hardest part of your local AI filmmaking workflow?

document fetched 2026-10-07 00:40 · canonical https://reddit.com/r/comfyui/comments/1wz9qy1/we_built_a_working_local_ai_filmmaking_pipeline