The open-source AI video studio

Your next film.
Made on your own GPU.

Generate shots with sound, recast whole casts, paint effects onto real footage, upscale, score and cut the final film, in one local workspace with open-weight models and native ComfyUI. Comfy Cloud is optional, studio by studio.

MADE ENTIRELY IN AFTERTAKE93 SECONDS · 72 SHOTS · ITS OWN SONG

Generated portals, castles, recast characters and finishing looks, cut to an open-weight ACE-Step song. Every shot, title and transition is an editable Film layer; the generated plates were upscaled on Comfy Cloud through Aftertake.Film MP4 ↧ · Editable projects ↗

Earlier 53-second demo ↧Its chapter guide + project ↗Native H3 · reference, tracked mask + result ↗Narrated cuts · 20-second native workflow ↗Follow voice · highlighted words ↗

From words to video · native LTX + sound ↗Two vertical ads · narrated hooks A + B ↗Fresh render: lighthouse to fantasy castle ↗Motion typography · 12-second sample ↗Change the cast · native H3 workflow ↗Local agent editing · MCP proof ↗

01 / MAKE IT

A sentence.
A shot with sound.

Text to video with sound on LTX-2.5, stills with Z-Image Turbo, camera moves from one frame and songs from ACE-Step 1.5. Start from 18 recipe starters, and see the time before every render.

02 / CHANGE IT

Your cast.
Your effects.

Track a person with SAM3 and recast them with MiniMax H3, then the next one in the same shot. Objects in front stay in front, with depth on the NPU. Paint a timed effect onto real footage, or transfer a performance.

03 / FINISH IT

Sharper. Smoother.
Cut to the beat.

Upscale with SeedVR2, smooth or slow motion with RIFE, grade with Color & finish, then cut with punch titles and transitions on detected beats. Agents get the whole studio through 45 MCP tools.

Start with something real

Recorded examples.
Ready to make your own.

The launcher checks your computer, installs what is missing and opens a workspace full of real results. Edit, finish and cut without a GPU. To generate, add a model profile for an NVIDIA GPU, or switch any of five studios to Comfy Cloud.

Quick start, model profiles and licenses ↗
NODE 22+ · GIT
git clone https://github.com/ItsSNOWMOBILE/aftertake.git
cd aftertake
node scripts/aftertake.mjs --examples

MIT app code · Official Comfy SDK · Local open-weight inference. Model licenses apply separately. The documented local renders used an RTX 5080 with 16 GB VRAM; optional Comfy Cloud rendering was validated end to end on a Standard plan. Generation quality varies by scene and prompt; reference consistency and camera guidance remain experimental.

Footage: Michael Burrows / Pexels (fictional coffee ad), Dean Diemert / Pexels, Florian Delée / Pexels, Curtis Adams / Pexels and Mikhail Nilov / Pexels. Silent bridge excerpt: Tears of Steel / Blender Foundation (CC BY 3.0), with generated character replacement. Procedural study: Aftertake contributors. Fonts: DM Sans and Manrope (OFL). Full attribution · Build evidence