video studio

turn your writing into an 88-second cinematic video

how the video pipeline works

every 88-second video is generated through a multi-stage AI pipeline that turns your stream-of-consciousness writing into a psychoanalytic film. the entire codebase is open source.

the pipeline

1 script generation — Claude reads your writing as a psychoanalytic director — reading for structure, repetition, avoidance, and breakthrough. it generates a story spine (the wound, the desire, a transformation arc, a controlling visual metaphor) then directs 8-15 scenes. the structure is free-form — the writing dictates the form.

2 sequential chain — each scene is generated one at a time: Gemini creates the image using Anky reference images + continuity frames extracted from the previous scene's completed video. then Grok generates the video clip from that image with directed sound. finally, ffmpeg extracts reference frames from the completed clip and passes them to the next scene. visual continuity is mechanical — each scene literally sees what came before.

3 stitching — ffmpeg concatenates all clips into a single 88-second video, then encodes at 1080p, 720p, and 360p.

the story spine

before generating scenes, the system finds:

the wound — what the writing orbits without landing on (from avoidances + patterns)
the desire — what the writing itself reaches toward (from breakthroughs + themes)
the arc — the psychic journey: "from X, through Y, to Z"
the visual metaphor — the dream-image the writing produces

the psychoanalytic reading

each scene includes an analyst's margin note — what the system sees in your writing that you might not see yourself. defenses, desires, repetition patterns, moments of genuine breakthrough. this is visible in the scene overlay when you click any frame.

memory system

if you've written before, the pipeline accesses your accumulated memory: psychological profile, recurring patterns, avoidances, breakthroughs, emotional signatures, and echoes from similar past sessions. this means your videos get more personal and insightful over time.

cost

$5.00 USDC per video. Grok charges $0.05/second of video (~$4.40 for 88s), plus Gemini images (~$0.40) and Claude script (~$0.05). paid via wallet transaction on Base.

open source

the entire pipeline is written in Rust (Axum + Tera + HTMX). you can read, fork, and modify every part of it:

github.com/jpfraneto/anky-monorepo

key files:

src/pipeline/video_gen.rs — the full pipeline: script generation, image gen, video gen, stitching
src/routes/api.rs — API endpoints for generation, polling, resume
src/services/grok.rs — Grok image-to-video integration
src/services/gemini.rs — Gemini image generation
src/memory/ — the memory system (embeddings, recall, profiles)
templates/video.html — this page you're looking at

sign in to create videos

write for 8 minutes to create an anky, then transform it into an 88-second video powered by Grok.