AI face swap tools in 2026: an 11-platform documentation audit

Quick answer

Ten of 11 selected platforms document both photo and video face swapping. We identified face-swap API references for four providers. Choose by the exact input, face mapping, export and billing route; these findings describe official documentation, not output quality or market share.

This edition adds AiFaceSwap.io, Vidnoz and Vidwud to our original eight-provider sample. Their official product and API pages were checked October 1, 2026; the first eight were checked September 28–29. Magic Hour publishes this audit and appears in the sample. We did not submit identical media to every provider, so these observations do not establish a quality winner.

What the 11-platform documentation audit found

10 of 11 selected providers document both photo and video face swapping. Pixlr remains the exception under our stricter inclusion rule: its dedicated tool documents photos, while its homepage advertises video without an end-to-end video path verified in this audit. We identified public face-swap API references for Magic Hour, Remaker, Akool and AiFaceSwap.io. API availability for the remaining seven was not systematically assessed; this is not a claim that they lack APIs.

The useful distinction is the route to a finished asset. A guest tool can offer a small daily allowance; another interface can show a credit cost at the button; another requires an account or subscription. Marketing labels such as “free” do not establish whether a user can download a usable, watermark-free output. Check the current product screen for the exact job before comparing plans.

Platform	Documented media	Access observation	Face-swap API reference
Magic Hour	Photo and video	Guest photo and video routes; separate daily allowances	Official API reference
Higgsfield	Photo and video	Photo and video apps listed; plan access not evaluated	Not assessed
Remaker AI	Photo and video	Video page displays a one-credit-per-second meter; photo route documented separately	Official API reference
DeepSwap	Photo and video	Upload flow for video, photo and GIF; checkout not evaluated	Not assessed
Pixlr	Photo tool; video advertised separately	Photo action shows two credits; video workflow not verified	Not assessed
Reface	Photo and video	Photo and video upload described; plan access not evaluated	Not assessed
Pica AI	Photo and video	Photo and video upload described; current allowance not evaluated	Not assessed
Akool	Photo and video	Face Swap and Live Faceswap documented; plan access not evaluated	Official API reference
AiFaceSwap.io	Photo and video	Photo, basic-video and advanced-video modes have separate allowances; checkout not tested	Official API reference
Vidnoz	Photo and video	Separate photo, video and multiple-face upload modes; account entitlement not tested	Not assessed
Vidwud	Photo and video	Photo and video/GIF upload fields shown; paid upgrade terms and final download not tested	Not assessed

The media and access columns describe what each vendor currently documents, not what we independently generated. The API column only names public face-swap documentation found in this audit. Open the linked official source for current plan details.

Pixlr source distinction: its dedicated Face Swap tool shows a photo upload and PNG download, while its homepage advertises Face Swap Video. We did not test that video path, so it is not counted as documented end-to-end video support here.

What this means for a buyer

For a one-off photo: start with the live upload and download flow. Magic Hour documents a no-signup guest photo route with a daily allowance and no watermark. Pixlr's photo swap button shows a two-credit action inside a broader editor. Those statements describe two access paths, not comparative image quality.

For video: test the actual length, resolution, watermark, face mapping, and export. Remaker's current video page displays a one-credit-per-second meter; its separate multi-face workflow matters when several people appear. DeepSwap's official upload page accepts video, photo, and GIF and describes choosing detected faces. The displayed checkout and result can differ by plan, region, and input.

For developers: compare the exact endpoint, not the consumer homepage. Published face-swap references in this sample include Magic Hour, Remaker, Akool, and AiFaceSwap.io. Check input URLs, face mapping, async status, output lifetime, retries and endpoint billing. A documented API is not evidence that every web feature is available through it.

API integration comparison: four documented routes

A face-swap API reference does not make the integrations interchangeable. We reread the four API references identified in this sample on October 10, 2026. These are documented input, mapping and result contracts; we did not execute the APIs or measure output quality.

Provider	Inputs and face mapping	Completion and output handling
Magic Hour	Video accepts a file URL/upload path or YouTube source, with start_seconds and end_seconds. Use all-faces for one replacement across detected faces, or individual-faces for up to five original_face/new_face mappings. The photo endpoint takes source_file_path for the replacement face and target_file_path for the base photo.	Read the returned project ID through Get Video Details or Get Image Details. Completed results provide download URLs with expires_at; save the accepted file before that URL expires.
Remaker AI	The photo API accepts binary target and replacement images. The video API takes a binary face image plus a reachable target-video URL; the documented video must be under 30 minutes and 500 MB. Multiple-photo mapping uses detected coordinates and ordered replacement crops in a ZIP.	Create a job, then poll its Fetch Job endpoint. The photo reference says generated images are deleted after 24 hours. The video reference repeats “images”; a video-output lifetime was not established by this wording.
Akool FaceSwap Plus	source_url supplies the replacement face; target_url supplies the image or video. Single-face mode simplifies the request; multi-face mode uses explicit source/target mappings. Detection-returned face crops need no bounding box; original-image URLs do.	Video runs asynchronously with a webhook or Get Result polling. The reference says API-generated resources are valid for seven days. Output availability is distinct from the provider’s complete privacy-retention policy.
AiFaceSwap.io	Video requires publicly reachable CDN URLs. Basic and multi-video routes differ; multi-video accepts reference-face crops and replacement URLs for up to three faces. Uploading and trimming are handled by the developer.	The reference describes an asynchronous task ID and callback webhook. No output-lifetime commitment was found in the checked page; confirm storage requirements before integrating.

Two implementation details that change the job

Separate browser pricing from API pricing. AiFaceSwap.io’s documented unenhanced multi-video formula is (face count + 1) × 2 API credits per second. Two mapped faces over ten seconds therefore require 60 generation credits, plus the two-credit face-extraction call when used. This is arithmetic from its API schedule, not a tested accepted-output cost or a dollar quote. A consumer-web meter cannot price that request.

Convert mapping coordinates correctly. Akool’s detection response uses [x, y, width, height]; its original-image mapping requires [x, y, x + width, y + height]. Cropped face URLs returned by detection skip that bounding-box step. This difference matters even when both routes are described as multi-face.

For a Magic Hour integration, start with the exact photo endpoint or video endpoint, then implement status and expiring-download handling before increasing job volume.

Supplement method: the eleven-provider media/access collection remains September 28–October 1. The four-provider API-contract subset was checked October 10, 2026. These four references were already identified in the sample; the other seven were not reassessed for API absence. This supplement used no accounts, paid jobs, output tests or checkout tests.

10-second AiFaceSwap.io API generation costs

API route	No enhancement	Enhance 1 (≤1080p)	Enhance 2 (≤2K)
Basic video	20 credits	40 credits	60 credits
Multi-video: 1 mapped face	40 credits	60 credits	80 credits
Multi-video: 2 mapped faces	60 credits	80 credits	100 credits
Multi-video: 3 mapped faces	80 credits	100 credits	120 credits

These are calculated generation charges for one ten-second request, using AiFaceSwap.io’s documented API schedule checked October 12, 2026. They exclude face extraction: add two credits per extraction call when used. The table contains twelve calculations for one provider, with no executed requests. Enhancement labels describe API settings; they do not prove delivered resolution or quality. Dollar cost, retries, refunds and accepted-output cost were not measured. Browser pricing is a separate schedule.

A better price comparison: cost per accepted result

Define one job before comparing price: a ten-second clip with two consenting people, a downloaded file at the required resolution, and stable identity through a turn. Record every attempt, displayed charge, rejected output, export restriction and manual correction. Divide total spend by accepted deliverables. This audit has not run a controlled 11-tool output test and therefore publishes no quality or accepted-output cost leaderboard.

Method and limits

Population: 11 deliberately selected browser platforms relevant to our photo and video face-swap readers. This is a purposive documentation sample, not a random sample or complete market census. Eight were selected in the first edition; AiFaceSwap.io, Vidnoz and Vidwud were added because they have distinct dedicated workflows and appear in our face-swap research or search-demand queue. The larger inventory suggests 16 primary-category entries and 59 feature-flag candidates, neither verified as a market count. Only the 11 checked providers are the denominator here. Both photo and video require a dedicated product page or interface description; a homepage claim alone is insufficient.

Collection window: September 28–October 1, 2026. For each selected provider we recorded dedicated photo/video documentation, a descriptive access observation and any face-swap API reference identified. Media support has 11 assessed providers; API absence has no assessed denominator, so unexamined cells say “Not assessed.” We did not create accounts, submit media, test checkout, measure quality or independently audit privacy. Current account and plan conditions may differ from public pages.

Local and open-source stacks are excluded from the 11-platform statistic because installation, hardware, component licenses and execution routes differ. For those, see our face-swap software guide. For a practical shortlist, see the eight-tool face-swap comparison.

Official source trail

Magic Hour: product or help page · face-swap API

Higgsfield: product or help page

Remaker AI: product or help page · face-swap API

DeepSwap: product or help page

Pixlr: product or help page

Reface: product or help page

Pica AI: product or help page

Akool: product or help page · face-swap API

AiFaceSwap.io: official product page · video workflow · face-swap API

Vidnoz: official product page

Vidwud: official product page

Three additions: what their official pages establish

AiFaceSwap.io: its separate photo and video routes describe uploads. Its API reference documents basic-video and multi-video operations. Browser and API credit schedules should be checked separately; one route’s quote cannot price the other.

Vidnoz: the dedicated face-swap page shows photo, video and multiple-face modes with original-media and replacement-face fields. This verifies a documented workflow, not the advertised speed, quality, free export or security claims.

Vidwud: the dedicated face-swap page shows separate photo and video/GIF input fields alongside paid upgrade controls. “Free” in a heading does not settle watermark, account or export conditions; inspect the final offer for your job.

What we will measure next

A stronger second edition will use the same consented source faces and target clips, fixed task definitions, recorded settings, every attempt and failure, observed charges, and representative downloadable outputs. The acceptance criteria will include identity stability, occlusion, multi-face mapping, resolution, and a usable export. Until that work is complete, this article remains a documentation audit.