Bria Image Enhance

Every photographer with a phone has a drawer full of images that are almost usable - soft focus, noise, JPEG mush. Bria Image Enhance is the "fix it in the cloud" version of that fix: it upscales and cleans up an image in one API call, returning a sharper, higher-resolution result as a normal ComfyUI IMAGE tensor. If you've wrestled with local upscalers and sharpening chains, this is the zero-setup alternative - and it's generative enhancement, not just an upscale, so it also restores detail the source never had.

It lives in the BRIA AI API nodes pack, which means the work happens on BRIA's servers (/v2/image/edit/enhance) and you need an API token from platform.bria.ai. The local cost is nil - no model download, no VRAM, no post-processing chain to tune.

The inputs that matter

• images and api_key - required. It batches, so feed it a whole folder of shots (via Bria Multi Image Select, for instance) and it enhances them one by one.
• resolution - the target size: 1MP (default), 2MP, or 4MP. This is the main lever: 1MP is a gentle cleanup, 4MP is a real upscale. Higher costs more, and output quality plateaus on badly degraded sources.
• steps_num - 10–50, default 20. More steps = more careful restoration at more cost. Default is a good place to start.
• preserve_alpha - on by default, so transparent images stay transparent instead of getting a background baked in. Leave it alone unless you want a solid background back.
• seed - default 681794, useful for consistency across a batch.

Outputs: output_images (the enhanced tensors) and seeds (the seeds actually used, as a string - handy for reproducibility).

Install

Standard pack install - no requirements file, no models:

cd path_to_comfyui/custom_nodes
git clone https://github.com/Bria-AI/ComfyUI-BRIA-API.git

or ComfyUI Manager → search "BRIA API" → restart, and drop your token in.

When to use it vs. local upscaling

Local upscalers (see the KB's upscaling essay) are still the move when you want control - specific scale factors, a model you trust, zero per-image cost. Enhance is the move when you want hands-off restoration and you're okay paying per image. The generative part is the differentiator: it doesn't just interpolate, it repairs, which is why a noisy phone shot comes back cleaner rather than just bigger. Where people get burned: expecting 4MP magic from a 200px thumbnail - enhancement is bounded by what's in the source - and forgetting that every run costs a credit, so batch with a locked seed instead of iterating one image at a time.