Open tools for
training and agents.
Trainers, research and 65 models from the Alvdansen studio. Built by artists, for artists.
Tools
Signet
BETAVideo LoRA trainer with reference control built in: single frame, keyframes, video to video.
View on GitHubFlimmer
VIDEO LORAFrom raw footage to a trained LoRA. Clips, captions, checks and training for Wan 2.1 and 2.2.
View on GitHubVellum
MCPProduction tracking and provenance for AI asset generation, as MCP tools.
View on GitHubGridwatch
EVALBuild comparison grids of samples, steps by prompts, for video or images.
View on GitHubComfyUI Template Agent
AGENTAn agent for building and checking ComfyUI templates.
View on GitHubResearch
Animating on Twos
Aug 2026Keyframe-animation adapters on a pretrained video model.
Full-shot anime generation
Oct 2026Which video models move convincingly, and a one-second repair.
Forgetting on Purpose
May 2026Generalization as the quality test for small-dataset LoRAs.
Qwen chained-edit method
House methodTraining one LoRA across Qwen-Image and Qwen-Image-Edit.