Runway Image to Video (Gen3a Turbo)

The whole point of this node is that it's the smallest possible runnable Runway workflow: one starting image, one prompt, one video out. No keyframes, no end frame, no camera rig. If you want to see what Runway's image-to-video feels like without reading a manual, this is the on-ramp.

It uses the Gen3a Turbo model, and there's an honest caveat that matters before you build a production pipeline on it: Runway deprecated Gen3a in 2026 and the node carries the is_deprecated flag in core. It still runs fine as of this writing, and it's cheap and fast relative to newer Runway models, but the label means Runway is steering people to Gen-4. If you found this node in a search, consider also glancing at the Gen-4 image-to-video node - it's the same shape, minus the deprecation.

What you actually set

The schema is refreshingly small:

• start_frame - the image that opens the clip. Runway animates out of this and everything after is generated. The node validates that it's within a 1:2 to 2:1 aspect ratio and under 8000px on a side, which covers virtually any image you'll feed it.
• prompt - what happens in the clip. Best practice here is to describe motion and scene changes, not restate the image. The still is already in the frame; the prompt's job is the animation.
• duration - Runway's duration dropdown (5 and 10 seconds, roughly).
• ratio - Gen3a Turbo only offers 768:1280 (portrait) and 1280:768 (landscape). No square, no ultrawide. Match it to your start frame or the model will letterbox or crop.
• seed - 0 for random, anything else to replay a take. For a paid API node this is more useful than it sounds: when a run costs you credits, being able to reproduce the good one instead of rerolling is real money.

The mechanism, quickly

ComfyUI uploads your start frame to its own proxy, submits it to Runway's image-to-video endpoint as a first-position prompt image, and then polls the task until the render is done - the code estimates around a minute, and it downloads the finished video straight back into the graph. That means it needs three things your local nodes don't: a Comfy account signed in on the frontend, credits on that account, and a working internet connection. No GPU, no model download, no VRAM at all. If you're used to the rest of ComfyUI, the "nothing is running locally" part is the biggest mental shift.

The single output is a VIDEO object. Wire it into a video save node to write the file, or straight into another node if you want to pass the clip on for editing, upscaling, or comparison.

Where it trips people up

The most common failure is assuming the API respects your aspect ratio choices. If your start frame is square and you pick the landscape ratio, Runway won't magically reframe it well - it'll generate around whatever it's given, and you'll wonder why the composition looks off. Crop the frame yourself before it goes in. Second gotcha: it's a paid call, so a typo in the prompt or a wrong ratio is a charged mistake. Cheap relative to the big Runway models, but not free.

If this is your first Runway node, it's the right one to start with. If you need an end frame too, the First-Last-Frame node adds that control for a modest price premium. If you need the current model, Gen-4 is the one to watch.