Face swap

I am writing to submit a feature request for a native face swap / identity-preservation capability within the Venice image tools.

Venice's current image editors are diffusion-based, which means they re-synthesize the face during an edit. The result is that facial identity tends to drift, and there is no mechanism to reliably preserve a source identity across an edit. The specific task of swapping one face onto another while keeping likeness intact is therefore not currently well served by the existing tools.

The proposal

Add a dedicated face-swap node built on the established open-source InsightFace pipeline:

Face detection and alignment (SCRFD / ArcFace embedding)

Identity transfer via the inswapper_128 model

Optional post-swap restoration (CodeFormer) for high-resolution targets

This is the same underlying stack that powers the most popular browser-based face-swap services, and it is purpose-built for exactly this job rather than being adapted from a general editor.

Why it fits Venice well

Content-neutral and NSFW-capable. inswapper_128 has no content restrictions, so it would work seamlessly on Venice's adult image and video outputs — a natural fit for the platform's positioning.

Cheap and fast. Inference runs in seconds and is far lighter than a diffusion pass, so it should be inexpensive to host.

Complements existing tools. It pairs naturally with the current generation and editing models without replacing them.

Suggested capabilities

Multi-face targeting within a single image

Adjustable blend mask (with neck/hairline padding) for clean seams on exposed skin

A pixel-boost / resolution option for high-resolution targets

One technical note

The inswapper_128.onnx model is free for personal use but requires a license for commercial deployment; the InsightFace code itself is MIT-licensed. Worth flagging early, though the team is likely already aware.

I am writing to submit a feature request for a native face swap / identity-preservation capability within the Venice image tools.

Venice's current image editors are diffusion-based, which means they re-synthesize the face during an edit. The result is that facial identity tends to drift, and there is no mechanism to reliably preserve a source identity across an edit. The specific task of swapping one face onto another while keeping likeness intact is therefore not currently well served by the existing tools.

The proposal

Add a dedicated face-swap node built on the established open-source InsightFace pipeline:

Face detection and alignment (SCRFD / ArcFace embedding)

Identity transfer via the inswapper_128 model

Optional post-swap restoration (CodeFormer) for high-resolution targets

This is the same underlying stack that powers the most popular browser-based face-swap services, and it is purpose-built for exactly this job rather than being adapted from a general editor.

Why it fits Venice well

Content-neutral and NSFW-capable. inswapper_128 has no content restrictions, so it would work seamlessly on Venice's adult image and video outputs — a natural fit for the platform's positioning.

Cheap and fast. Inference runs in seconds and is far lighter than a diffusion pass, so it should be inexpensive to host.

Complements existing tools. It pairs naturally with the current generation and editing models without replacing them.

Suggested capabilities

Multi-face targeting within a single image

Adjustable blend mask (with neck/hairline padding) for clean seams on exposed skin

A pixel-boost / resolution option for high-resolution targets

One technical note

The inswapper_128.onnx model is free for personal use but requires a license for commercial deployment; the InsightFace code itself is MIT-licensed. Worth flagging early, though the team is likely already aware.

Face swap is consistently one of the most in-demand use cases in this space, and Venice currently has no native answer. My sense is that a simple, reliable face-swap feature would see v

Face swap is consistently one of the most in-demand use cases in this space, and Venice currently has no native answer. My sense is that a simple, reliable face-swap feature would see very strong adoption.