imageToVideo

imageToVideo is the local control surface for the diffusion pipelines I run on my own hardware. No cloud, no API keys, no per-frame billing. The model loads, the queue runs, the output writes to disk. That is the whole loop.

GPU routing is the interesting part. The workstation has two RTX 5060 Ti cards plus an RTX 4080 reserved for gaming. The app pins generation to the two 5060 Ti cards, watches VRAM headroom per card, and dispatches the next batch to whichever has room. The 4080 stays untouched so a long render does not steal frames from anything else.

The frontend is Gradio — fast to iterate, easy to expose over LAN. Prompt presets cover common image-to-video framings (slow zoom, parallax pan, ambient drift). The batch queue lets a long session run unattended; the dashboard shows per-job progress, ETA, and which card is doing the work.

Used daily as the production tool behind a Discord bot fleet. Generation requests come in over webhook, hit the queue, and write back to the requesting channel when finished. Boring infrastructure, interesting outputs.