FrameFamily

A modular AI ecosystem focused on frame-based image generation, training, and visualization.

Components

• Training
  FrameForge
  AI training, dataset preparation, and orchestration within the Frame ecosystem.

• Training (Next Gen)
  FrameForge2
  Successor to FrameForge with an in-house trainer and improved pipeline reporting integration.

• Viewing
  FrameView
  Visualization, inspection, and analysis of generated frames and training results.

• Generating
  FrameCreate
  Generative image AI of the Frame ecosystem.
  Work in Progress

FrameForge2

End-to-end dataset and LoRA training pipeline that turns ZIP uploads into clean, tagged image datasets with a web-first workflow. Best use in combination with Grok Imagine Image to Video.

FrameForge2 is the successor to FrameForge. It replaces the former Kohya-based trainer with an in-house trainer and adds tighter pipeline integration with reporting.

FrameForge2 is not production-ready yet and still needs significant refinement. Training is possible, but results do not meet quality targets.

FrameForge2 is designed for fictional, stylized, and synthetic content.
Use on real individuals without consent is explicitly discouraged.

Important Notice: FrameForge2 is currently only for PonyXL training calibrated. More will follow if demand comes.

Features

• Pipeline stages: import → select → crop → tag → train → package.
• Dataset flow: ZIP ingest → capping → selection → crop/flip → autotag → manual tag editing → finalize dataset.
• Training flow: profile-driven configs → run monitoring → LoRA packaging + sample previews.
• Orchestrated services: initiator, orchestrator, finisher, DB broker, webapp.
• Web UI for uploads, queue monitoring, manual tagging, downloads, and train profile editor.
• Queue management with reorderable queued runs.
• Preview generation and result browsing in the UI.
• AutoChar presets with online selection and filtering.
• System status view with service health and progress signals.
• Structured error logging with per-run details in the UI.
• Training with an in-house trainer (no external wrappers).

Support and Questions -> Discord https://discord.gg/TB5DHMNa5J

Quick Start

./scripts/setup_all.sh

Open the Web UI at http://localhost:3030.

Prerequisites: Python 3, Node.js + npm, ffmpeg in PATH. GPU recommended for tagging; required for training.

Configuration

Variable	Default	Notes
DB_BROKER_URL	(set by systemd)	Required for webapp DB access.
PORT	3030	Web UI port.
HOST	0.0.0.0	Web UI bind address.

Usage

./.venv/bin/python workflow.py --autotag --gpu

Add --train to run LoRA training.

For UI usage and workflows, see the docs:

• insite-docs/quickstart.md
• insite-docs/ui.md
• insite-docs/workflow.md

Development

• Setup (all-in-one): ./scripts/setup_all.sh
• Services: systemd/frameforge-*.service

Security

FrameForge2 is intended for local use only. Do not expose it to the public internet.

License

MIT