Show HN: Making 10-minute AI movies with Claude Code and Higgsfield ($200 each) A developer has built a pipeline using Claude Code and Higgsfield's CLI to produce 10-minute AI movies for about $200 each, with Seedance 2.0 for video, Nano Banana Pro for images, Sonilo for music, ElevenLabs for voices, and Gemini for quality control. The pipeline has generated at least three finished films, including an 11-minute comedy called 'The Long Game,' and the full playbook is documented in MOVIE_LESSONS.md. A working pipeline for making short films 5–12 minutes, dozens of shots, consistent characters and voices with Seedance 2.0 video , Nano Banana Pro images , Sonilo music — all three via Higgsfield's CLI — plus ElevenLabs voices , Gemini audio/image QC , and a lot of ffmpeg. About ten films have been made this way; the first complete one, The Long Game an 11-minute comedy, ~80 shots, ~15 iteration passes , lives in long game/ as the worked example. A taste of the output: The Long Game's cast, voices, and start frames https://dawndrain.github.io/movie-gen/long game/ , the storyboard https://dawndrain.github.io/movie-gen/long game/storyboard.html , the voice auditions https://dawndrain.github.io/movie-gen/long game/auditions.html — and the finished film https://www.youtube.com/watch?v=q40M08SOhGs . Two more finished films: Homo Sapien https://www.youtube.com/watch?v=QlP-dhZTQUE the music video and Walter's Deal https://www.youtube.com/watch?v=pHtF ApUjRo . There's also a blog post about the whole endeavor: Making AI Movies https://dawndrain.notion.site/Making-AI-Movies-39cbb81dad798077b43adcd96e3ebe75 . Start with MOVIE LESSONS.md — the playbook: the pipeline, the laws retakes regress; after two failed rewordings change what happens; the problem dictates the fix , prompting, voices, sound, and costs. The dated per-project postmortems behind those rules live in PROJECT LOG.md /dawndrain/movie-gen/blob/main/PROJECT LOG.md . Anchors character portraits → start frames one per scene, anchors as refs → cast voices and TTS the script → animatic the whole film as stills + TTS, ~free — iterate here → Seedance clips start frame + anchors + voice refs → music and ambience as assembler layers → ffmpeg assembly → storyboard.html + preview → director notes → cheapest fix per note → repeat. gen.py submit Seedance / Nano Banana jobs handles the 8-job cap tools/ pool run.py batch runner; skips finished shots, so re-run = retry assemble.py spec-driven film assembler cuts, music spans, ambience beds, grades, fades — see long game/film spec.py dub clip.py replace a line's audio without re-rendering pitch check.py median-F0 screen for wrong-voice takes listen.py Gemini listens to audio for QC fault-finding prompts only templates/ best-of-breed per-film tools to copy into a new project: animatic, auditions, ambience, images, storyboard, batch emitter, frame edits, upscale, dub pass long game/ the worked example: story, film spec.py, storyboard gen, archive/ of every iteration script media is gitignored other movies/ the other film projects sagas, Donner Party, Walter's Deal, the Homo Sapien music video, ... . Copyrighted source texts and a few fan-IP projects are local-only via .gitignore. veo3 compare/ the std-vs-fast blind test harness + prompts Prerequisites: Python 3.10+, ffmpeg / ffprobe on PATH, Node 18+ for the Higgsfield CLI npm i -g @higgsfield/cli — the only path to Seedance 2.0 / Nano Banana Pro , and pip install numpy pitch check . Optional per tool: edge-tts free draft TTS , demucs + Whisper dub QC . npm i -g @higgsfield/cli , then higgsfield auth login && higgsfield workspace set