Bria Video Increase Resolution

Video upscaling is the hardest of the three upscaling jobs, because every frame has to agree with its neighbors - upscale them independently and you get shimmer and breathing. Bria's video upscaler handles that coherence server-side on Runware's GPUs, and the node itself is gloriously simple: a video in, an upscale factor, and an output format. If you've generated a video with another Runware node and it's too soft, or you've got footage at a URL that needs to be bigger, this is the node.

Inputs, and the surprise in the output

• video - required STRING. UUID or URL, same convention as every video node in this pack. A Runware generation returns a UUID you can feed straight in; anything else needs to be at a hosted URL. Do not expect a VIDEO socket.
• upscaleFactor - 2 or 4. The whole show. 4 quadruples the pixels and costs more per run and takes longer; 2 is usually the sane choice unless you're going from genuinely small to large.
• outputFormat - MP4/WEBM/MOV, default MP4. outputQuality (20–99) and ttl - leave alone.

Here's the part the name doesn't warn you about: the output is IMAGE, not VIDEO. This node's task type is upscale, which the pack routes through its image-producing path - so what you get back is a batch of upscaled frames as an IMAGE tensor, not a single video file. Preview them, save them, or re-combine into an actual video if that's your end goal. It's a quirk of the pack's task-type routing (compare Runware_bria_video_eraser, which returns native VIDEO because it's a videoInference task), and it's the #1 "wait, where's my video" moment with this node.

Install and key

Standard pack setup, once for everything:

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

Restart, add your Runware API key (Settings → Runware API key, or RUNWARE_API_KEY). Nothing downloads - inference happens on Runware's GPUs and the run cost prints on the node's title bar.

The honest take

For a one-off "make this clip sharper" job, this is the easiest option in ComfyUI, period - no model files, no tiled-diffusion scaffolding, no VRAM ceiling. The two gotchas are the string video input (UUID/URL, not a tensor) and the IMAGE-batch output, which trips everyone the first time. Also worth remembering, from the wider upscaling discourse: generative upscaling rewrites detail, so if the video has a face you need to stay recognisable, don't expect a 4x pass to leave it untouched. And since there's no audio knob on this node, don't count on it preserving a soundtrack the way the Video Eraser does - if the audio matters, keep the original around and re-mux.