Beeble SwitchX Image Edit

Most "edit this image" models regenerate the whole frame and hope the subject survives. Beeble SwitchX inverts the priority: it swaps the scene - background, lighting, costume - while preserving the original subject's pixels. That's a different contract from Kontext-style editing, and it's the one you want when the person or product in the shot is the asset and everything around them is negotiable. The node shipped into ComfyUI core in May 2026 as part of the Beeble partner-node pair.

Partner-node preamble, quickly: this is a paid remote call through Comfy's proxy at api.comfy.org to Beeble's SwitchX API, billed from your Comfy account credits, with nothing running on your GPU and no model files to download. The input image is downscaled to a ~2.77MP cap before upload (the description says "Max ~2.77MP" - that's the node doing the downscaling for you), and it's uploaded as PNG.

Inputs:

• image - the source. prompt (max 2000 chars) - the new look. You need at least one of prompt or reference_image (optional) to describe the target scene; the node refuses to run with both empty. A reference image whose look you want to adopt plus a prompt usually beats either alone.
• alpha_mode - the heart of the node. A DynamicCombo with four options: auto isolates the main subject automatically, fill regenerates the entire frame while preserving geometry, select uses a first-frame keyframe to pick the subject, and custom uses a per-frame alpha matte you provide. Picking select or custom reveals the nested mask input you need to wire in.
• max_resolution - 1080p or 720p. The description in the source is blunt about what this means: higher resolution costs more credits.
• seed - rerun trigger only; results are nondeterministic.

Outputs: image (the edited result) and alpha - the alpha matte Beeble actually used, which is a genuinely useful output for compositing. It's empty in fill mode, which has no separate matte.

The grounded reality check: SwitchX is built for the "put my product/person in a new context" job - studio shots onto lifestyle backgrounds, characters into new scenes - and the subject-preservation guarantee is its whole selling point. The fill mode's "preserve geometry" framing is honest about the trade: it can regenerate everything, including the subject's surroundings and even the subject's own pixels, so auto is where most people should start. And because the output includes the alpha matte, you can take the switched scene, keep the original subject cutout, and composite - which is the pro move and the reason the alpha output exists.

Costs scale with resolution and the run count; check the badge before you batch. If you're doing character consistency where the identity must not drift, this is a strong alternative to fighting LoRAs.