AI Movie Studio V2 ComfyUI Tutorial: Install, Connect and Build a Scene

AI Overview

What is AI Movie Studio 2?

It is an open-source filmmaking workspace for organizing scenes, references and shots while sending generation jobs to local ComfyUI or cloud providers.

Do I need ComfyUI to use AI Movie Studio 2?

You need ComfyUI for local generation. Cloud-only drivers can work without a local GPU, but they require the relevant provider credentials.

What ports does the local setup use?

The defaults are ComfyUI on 8188, the backend on 8001 and the frontend on 3000. Verify each service before troubleshooting generation.

Can I import my own ComfyUI workflow?

Yes. Export API JSON, register it in Settings, then resolve any missing checkpoints, LoRAs, VAEs or encoders before running it.

What AI Movie Studio 2 Adds to ComfyUI

This AI Movie Studio V2 ComfyUI tutorial starts with the distinction that prevents most confusion: ComfyUI remains the generation engine, while AI Movie Studio 2 acts as a filmmaking workstation around it. Instead of rebuilding one giant graph for an entire project, you organize a project into scenes, references and shots, then send specific image or video jobs through a driver.

The current project describes a browser-based, model-agnostic workflow. Its tools include a 3D camera stage, shot presets, multi-reference generation, prompt variations, retakes, LoRA selection, a timeline, dialogue and audio, plus XML export for an editor. The interface can call local ComfyUI or supported cloud APIs. Because the project is under active development, treat the README and Settings panel in your installed revision as the source of truth; names and experimental features can change quickly.

This structure is useful when your real problem is continuity rather than a single beautiful frame. A scene recipe can hold the character, location and prop references. A shot can record the chosen angle and generation recipe. A rejected take can be rerun without erasing the accepted storyboard around it.

An original finished-output concept for this guide. The red coat, silver case, motel, vintage car, rain and oversized moon form a continuity recipe that can be checked across shots.

Before You Install

Choose local, remote or cloud generation

Local ComfyUI gives you direct model and node control, but you maintain the GPU environment, model files and custom nodes. A remote ComfyUI server keeps that graph control on another machine. Cloud drivers reduce installation work but introduce provider pricing, credentials and service-specific limits. The local-versus-cloud AI video guide helps frame that operational choice before you invest in a setup.

For the current manual installation, plan for Python 3.10 or newer, Node.js 18 or newer and a running ComfyUI instance. The upstream hardware guidance positions 8 GB VRAM as a minimum tier for image work and short clips, 12–16 GB as a broader working range, and 24 GB or more for heavier multi-reference or batch use. Those are project recommendations, not guarantees: your model, resolution, frame count and quantization determine actual memory use.

Keep the first test intentionally small

Prepare one character portrait, one location reference and one simple image workflow that already runs directly in ComfyUI. Do not start with a long take, multiple LoRAs, voice cloning and a new video model simultaneously. A small known-good job tells you whether the connection and driver mapping work before creative complexity enters the diagnosis.

AI Movie Studio V2 ComfyUI Tutorial

1. Start ComfyUI and verify port 8188

Launch ComfyUI in its own terminal and confirm its browser interface opens. The default AI Movie Studio configuration expects http://127.0.0.1:8188. If your server uses another address, a reverse proxy or authentication, record that before configuring the workstation. Do not expose an unauthenticated ComfyUI instance to the public internet merely to make the connection easier.

cd /path/to/ComfyUI
python main.py

Queue the small known-good workflow inside ComfyUI first. If that fails, repair ComfyUI before involving AI Movie Studio 2.

2. Install and start the backend

From the repository's backend directory, install the Python requirements. The current manual start command launches the API on port 8001; open its health endpoint and require a successful response before moving on.

cd backend
pip install -r requirements.txt
python main.py serve --reload

Keep provider secrets out of screenshots and committed files. Local ComfyUI works without cloud keys. If you later add a supported provider in Settings, confirm what is stored locally and restart the backend when the current build requires it.

3. Install and start the frontend

Use a third terminal for the frontend. After dependencies install, the development server normally opens on port 3000 and proxies requests to the backend.

cd frontend
npm install
npm run dev

Create a project only after the browser loads and the backend health check passes. If the interface opens but every generation fails instantly, inspect the backend-to-ComfyUI connection rather than reinstalling the frontend.

4. Use Docker when repeatability matters

The Docker route starts the application stack with one command, but it does not containerize ComfyUI in the current project design. ComfyUI still runs on the host, usually on 8188, and the backend container reaches it through host.docker.internal. Both application services bind to localhost by default and project data persists through bind mounts.

docker compose up --build

Docker reduces Python and Node setup drift. It does not solve missing model files, an unhealthy ComfyUI server or a workflow that references unavailable nodes.

A successful setup should produce a new composition without losing the character, red coat, case, car, moon or lighting logic.

Register a Custom ComfyUI Workflow

Export API JSON, not the visual workflow file

Build and test the graph directly in ComfyUI, then save it in API format. The execution JSON is what an external application can queue. Use a clean workflow with obvious input points for prompt, seed, width, height and source image. A graph that depends on hidden local state is harder for any driver to control reliably.

Open Settings in AI Movie Studio 2, find Custom ComfyUI Workflows and add the exported JSON. Give it a stable display name, driver ID and image or video category. The current workflow analyzer scans references to checkpoints, LoRAs, VAEs, CLIP or text encoders, UNets, ControlNets and upscalers, then compares them with the connected ComfyUI instance.

Resolve models before testing parameters

A missing-model warning is different from a generation-quality problem. Put each file in the directory expected by its node, refresh the model list and confirm the warning clears. Only then tune steps, CFG, seed or LoRA strength. This order prevents a simple path mismatch from becoming hours of prompt experimentation.

When the workflow appears in the model selector, run the same small reference and prompt used in the direct ComfyUI test. Compare output dimensions, seed behavior and saved files. If the results differ, inspect which widget values the driver overrides and which remain at the JSON defaults.

The reverse angle adds new information while keeping the key, silver case, doorway, car and moon spatially readable.

Build a Consistent Multi-Shot Sequence

Create the scene recipe before the shot list

For the motel example, define the woman, red coat, silver case, clerk, reception counter, vintage car, rain, amber practicals and oversized moon once. Assign every reference a role. Then plan three shots: exterior arrival, counter exchange and return to the car. The camera can change, but the recipe tells reviewers what must survive.

Generate storyboard frames before video. Use a wide frame to establish geography, a medium frame for the case handoff and a closing wide shot that restores the exterior. If a frame changes the actor or relocates the car, reject it before spending on motion. The multi-model AI video workflow explains how one acceptance checklist can survive when different shots use different engines.

Finished three-shot environment-continuity sample

An existing generated output from our media library, not an AI Movie Studio 2 benchmark. Inspect whether architecture, light and subject placement remain coherent across shots.

Review motion and save the generation recipe

Turn only accepted frames into clips. Watch each result at normal speed for face changes, sliding feet, prop deformation, camera reversals and abrupt lighting shifts. Save the seed, prompt, workflow ID, model choices and LoRA strengths with the shot. Without that recipe, “retake” becomes a new experiment instead of a controlled correction.

For repeated work, batch only after one representative shot passes. The ComfyUI batch-processing guide shows how to preserve an input manifest, isolate failures and keep outputs traceable rather than trusting a long unattended queue.

The closing frame proves continuity at a wider distance: the character silhouette, prop, clerk, car, motel and moon all remain recognizable.

When Seedance Agent Is the Faster Route

AI Movie Studio 2 plus ComfyUI is attractive when you want local control, custom nodes, workflow JSON, specific model files and direct parameter access. That control also means you own installation, updates, driver mapping, missing models, GPU capacity and failure recovery. For a technical creator, the maintenance cost may be worthwhile. For a marketing team that mainly needs an approved sequence, it can become a second production job.

Seedance Agent approaches the same filmmaking task from the brief and review layer. Provide the audience, concept, references, aspect ratio, duration and acceptance rules; review the proposed script and shot plan before paid generation; then reject or rerun only the weak shot. You can still use the text-to-video generator or image-to-video generator for direct tests, but the agent keeps the references, decisions and review trail together.

Seedance Agent finished product-ad output

This existing Seedance Agent output demonstrates the alternative deliverable: a reviewable moving result rather than a local setup screen.

Choose based on the work you want to own. Use the local stack when node-level experimentation is the goal. Use a hosted agent workflow when faster planning, approvals, continuity checks and selective reruns matter more than managing infrastructure.

Conclusion

A reliable AI Movie Studio V2 ComfyUI setup starts with three healthy services, one known-good workflow and one small reference test. Connect ComfyUI before adding creative complexity, register API-format workflows only after they run directly, resolve missing models before tuning parameters, and preserve a generation recipe for every accepted shot. When you want the same scene-planning and review discipline without maintaining the local stack, start your next sequence with Seedance Agent.