JAP JAP

AImation

Workflow for audiovisual production with AI

$ git clone github.com/JoseAPortillo/AImotion

$ cat README.md

AImation is a desktop-first node workflow for generating animations and video from image, video, and prompt inputs. You build a graph of connected nodes — an input video, a prompt, a generation step, sampling parameters — and the engine renders a new animation out of it.

Local generation runs through diffusers; cloud adapters cover Kling, Runway, and Seedance. The shell is Tauri (local file access), the editor is React + React Flow, and the brain is a FastAPI sidecar in Python.

$ ls media/

Node editor — Starting app
Cloud adapters generation — Runway render with Gen-4 model
Cloud adapters — Kling / Runway / Seedance output
Video1 — text to video
Video2 — image-2-image
Video3 — video-2-video

$ ls src/

Tauri React React Flow FastAPI diffusers Transformers GGUF LTX-Video Wan 2.2 CogVideoX Kling Runway Seedance ControlNet LoRA LLM VLM

$ node --list

inputImage Input
inputVideo Input
inputPrompt Input
inputAudio Input
coreGeneration
coreSampling Parameters
coreAspect & Resolution
coreOutput

Plus generator and adapter nodes: Text-to-Image, Image-to-Video, Video-to-Video, LLM generator, VLM, ControlNet, LoRA, and CV task processors.

$ cat generation.md

Local — diffusers runners

  • LTX-Video 2B (phase 1 target)
  • Wan 2.2
  • CogVideoX
  • GGUF quantized models
  • FP8 + CPU-offloaded text encoder

Cloud — API adapters

  • Kling
  • Runway
  • Seedance
  • Credential manager
  • Credit manager

Phase 1 runs LTX-Video 2B locally on an RTX 3080 with VRAM management: center-crop + resize, flow-matching denoising loop, and async task queue with progress polling.

$ cat architecture.md

  • src/backend — FastAPI sidecar: API routes, config, models, generator runners, task queue, model registry, credit & credential managers
  • src/frontend — React + React Flow node editor
  • providers — cloud adapters + local runners
  • workers — background execution
  • specs — SpecSecOps phase specs

$ cat methodology.md

Built with SpecSecOps — Spec-Driven Development with DevSecOps integrated from day one. Every change follows Explore → Propose → Spec → Design → Tasks → Apply → Verify → Archive, with adversarial review on every PR and tests as a hard requirement.

  • Desktop-first. Tauri shell, local file access, sidecar backend.
  • Specs before code. No implementation without an approved spec.
  • Tests are mandatory. No test == incomplete task.
  • Adversarial review before merge. Every PR gets a Risk + Reliability review.
  • Security in every phase. Not a separate step.

$ git status

status: Phase 1 — LTX-Video 2B local generation on RTX 3080: backend API, async task queue, progress polling, and full pytest suite.