SceneGen - Movies (In Development)

If you've read the SceneGen - MusicVideo article, you already know this node. SceneGen - Movies (In Development) is the same engine with a different menu label: open the right-click menu and you'll find it under SceneGen → Movies → SceneGen - Movies, while its twin lives under SceneGen → MusicVideo. Look at the inputs side by side and you'll find they're identical, down to the last slider.

So why two nodes? The author's framing in the README is that MusicVideo is for syncing visuals to beats and lyrics, while Movies targets "narrative storytelling, dialogue, and script-based content." The honest read: the label is the roadmap, and the code is the same 11-stage Gemini + Replicate pipeline you get with the music video node. The one meaningful difference is that the Movies variant is explicitly marked "In Development," which is your cue not to expect the movie workflow to feel finished yet.

What's actually different

Not the inputs. You'll find the full set here - audio, gemini_api_key, replicate_api_token, prompt_instruction, render_mode, creativity, dynamicity, word_influence, the use_* video model toggles, and so on. The outputs match too: the four image tensors (Environment Images, Asset Images, Actor Images, Scene Start Frames), the S1–S11 JSON audit trail, Cost Data (JSON), and Final Video Path.

The only field that hints at the narrative ambition is dialogues_gen. When enabled, the node asks Gemini to write dialogue for supported models (Veo 3.1 and Wan 2.5) and prevents those dialogue shots from being trimmed on beat, which would wreck the pacing of a scene. That's the one real "movies" behavior in the pack, and it's on by default in neither node - you have to flip it.

Everything else - the two API keys, the cost math, the render_mode trick for free testing, the model toggles that all default to off except Hailuo Fast - is identical to the MusicVideo node, so I won't repeat it here. If you're about to queue this for the first time, read that article's install and troubleshooting sections; they apply word for word.

My take

As of today, this node is a placeholder for a promise. The pipeline genuinely works - it's the same one that produces finished, beat-synced music videos - but "movies" as a distinct capability (real dialogue, scene-level continuity, narrative structure that survives an aggressive edit) isn't there yet. If you want to make a music video, use it or its twin; if you want to make a movie, this won't do it for you yet. Don't let the label sell you on a finished feature.

Worth saying: this is an obscure, single-dev pack with almost no community footprint, so the maintenance risk is real. If you build a workflow on dialogues_gen, check the repo for updates before assuming a bug is yours.

Install

Same as the pack's other nodes. ComfyUI Manager → search "Scene Gen", or:

cd ComfyUI/custom_nodes
git clone https://github.com/lazniak/scene_gen.git
cd scene_gen
pip install -r requirements.txt

Restart ComfyUI, then paste your Google Gemini key and Replicate token into the node fields. No model downloads - all inference happens in the cloud.