Goal (Jon, 2026-11-17: "yes, I want us to be able to ask a runner to make images/videos"): from Factiii, a person asks their runner for an image or a video, the runner makes it locally, and the result comes back to them. Ollama can't do this, so the runner needs a second local engine.
Two different things, don't mix them up:- Understanding images and video (reading a screenshot, describing a clip): Ollama does this today. Notes below, no new engine needed.
- Making images and video: not in Ollama. This card.
Making images and video (the work)
What Jon found (2026-11-17): generation needs ComfyUI.
- Video: Wan 2.2 (Apache 2.0, best open video, the 5B model fits 8GB with offload) or LTX-2.5 (video+audio, free under $10M revenue).
- MiniMax H3 is strongest but US-blocked.
- Image: FLUX via ComfyUI.
- Ollama may add diffusion later but not now.