Runway Aleph

Runway is the New York company that - in a twist it would probably rather forget - shipped Stable Diffusion 1.5 over Stability's objection in 2022 and then deleted its whole open-weights history in 2024. The company today is closed, commercial, and all-in on video and world models. Aleph is its flagship video generation model, and this node puts it behind a ComfyUI interface: you give it a source video and a prompt, and it re-renders that video into something new.

The format matters: this is video-to-video, not text-to-video. The required video input is an existing clip (a UUID or URL - Runware handles those, not IMAGE tensors). You restyle it, transform it, or push it in a new direction with the prompt. If you're generating video from scratch, that's a different node - this one is for taking footage you have and making it yours.

How it works

video + positivePrompt go to runway:2@1 with taskType: videoInference. The returned video is downloaded to your output folder and wrapped as a native video (VIDEO) output - or a file path on older ComfyUI where the native type isn't available.

Inputs that matter:

• video - the source clip, as a UUID or URL.
• positivePrompt - what you want done to it. Restyle, re-light, change the subject, add motion.
• referenceImages - optional IMAGE socket for character/identity anchoring.
• width / height - 1024×1024 default output canvas.
• seed - reproducible runs, 0 for random.
• numberResults - 1–4 candidates.
• providerSettings.runway.contentModeration.publicFigureThreshold - auto or low; Runway filters depictions of public figures and this controls strictness. Worth lowering if you're getting blocked on someone recognizable.
• safety.mode - none / fast / full for video safety checking. full is slower but more thorough.
• outputFormat - MP4 / WEBM / MOV.

Output: video (VIDEO).

Installing

Pack-wide, one install: ComfyUI Manager → Custom Nodes Manager → search Runware → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

No model downloads, light deps (runware-sdk, Pillow, soundfile). API key from the Runware dashboard.

Gotchas

Video is the most expensive thing in this pack, and video-to-video multiplies it - every queue is a full model run on a long clip. Check the title bar cost before you iterate 30 times. The video input being a UUID/URL rather than an IMAGE is a real friction point: you'll often have to upload a clip first (the pack's Runware Upload Image / related tools, or Runware's dashboard) and paste the ID. And if you're hitting content blocks on faces that look like anyone, adjust the publicFigureThreshold and the safety.mode rather than fighting the prompt.