Replace someone with yourself in a video with AI: Recast on Sume
Put yourself into an existing clip: send a public source video and one photo of you to H3 Max Recast on Sume. Limits, cost per clip and consent rules.

To replace someone in a video with yourself, send H3 Max Recast one photo of you and a public link to a 5 to 30 second clip. On Sume that is model: h3-max-recast on POST /v1/video-router/generate, with your photo as the only entry in reference_image_urls. The output keeps the source motion, camera, cuts and audio, with you in place of the person who was there.
It costs $0.375 a second at 768p and $0.5625 at 1080p on Sume, so a 10 second clip is $3.75 or $5.63. Two things are not on that bill: the clip has to be one you may reuse, and the person you replace is a real person too.
What do I need before I start?
Four inputs, all public HTTPS URLs, none signed or private:
- A source video of 5 to 30 seconds with no single shot longer than 15 seconds (a fal limit).
- One clear photo of you. Neither vendor publishes framing guidance, so use a sharp, well-lit one of you alone as a practical default.
- A resolution: 768p (Sume's default) or 1080p.
- The inspected source length in whole seconds, rounded up, for the
durationfield.ffprobereads it.
How do I submit it?
This call takes URLs, not uploads, so put the photo and clip at public HTTPS addresses first. Signed and private URLs and localhost are refused before any paid work. Sume's media inputs page lists the rules.
The call returns a job id immediately in async mode. Keep it, poll /v1/jobs/{id}/status, and read /result once the job is completed.
curl -X POST https://api.sume.com/v1/video-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: recast-me-001" \
-d '{
"model": "h3-max-recast",
"video_url": "https://example.com/original.mp4",
"reference_image_urls": ["https://example.com/me.png"],
"resolution": "768p",
"duration": 10,
"mode": "async"
}'What stays the same, and what changes?
From fal's description and Sume's catalog entry, read 2026-10-03:
- The soundtrack is the original one, so your face will sit on someone else's voice. See what Recast does with audio before you plan a voiceover.
- Sume has no seed on this model; a second try is a new take at the same per-second price.
| Element | Result | Source |
|---|---|---|
| Motion and camera | Preserved | fal model page |
| Cuts and edits | Preserved | fal model page |
| Audio | Source audio retained | fal API schema |
| The person | Replaced by your photo's person | fal and Sume catalog |
| Aspect ratio | Not settable | Sume rejects aspect_ratio |
| Output length | Same as source | Sume catalog |
What does a first test cost?
Do not start with the full clip. A 5 second slice is the shortest valid source and bills $1.88 at 768p, or $2.81 at 1080p (exactly $1.875 and $2.8125). That is enough to see whether your photo carries over to the person on screen. If the person wears a hat, turns away or sits in shadow in the slice, expect that to show in the result.
Then scale: the full 10 second clip is $3.75 at 768p, a 20 second clip $7.50 and the 30 second maximum $11.25. Move to 1080p only for the take you plan to publish.
How do I wait for it?
Recast is a queued job, not a request that stays open. Sume's sync and subscribe modes block for at most 30 seconds and then hand back the current state, so submit with async and poll, or use mode: webhook with a public HTTPS webhook_url. Either way, never resubmit just because your own process timed out: the first job is still running and still bills. Read the video only once the job is completed, from the result endpoint.
Whose clip is it, and who agreed?
Replacing yourself in someone else's video touches two permissions. Your own photo is the easy one. The clip is the other: if it came from a creator, a studio or a stock library, you need the rights to edit and republish it, and the person you are replacing is shown being replaced. Sume's Terms of Service say you keep ownership of content you submit and that you represent you have all rights and permissions needed, including permission to use any person's likeness or voice.
fal's page says commercial use is permitted, which speaks to its own terms and not to your rights in the footage. Treat that line as the vendor's side of the licence only.
Do I have to label the result?
Usually yes when the clip will be published. A real person made to appear in footage they were not filmed in is the textbook case platforms ask you to disclose. Rules differ by platform and by country, and they change. Sume's earlier posts cover TikTok's face swap disclosure and YouTube's disclosure for face-swapped video. Sume's docs do not describe any label that Recast adds to its output, so add your own where a platform or law asks.
Sources
Related posts
More in Use cases
- Roundel video ads in :06, :15 and :30: cut one master with Sume
Roundel's guides list :06, :15 and :30 video creative. Cut three lengths from one master with Sume video trim; what its pages leave to a Roundel rep.
- Shopify 3D model or 360 spin video: what Sume can make
Shopify's 3D slot needs a GLB or USDZ file. Sume makes video, not 3D models. Shopify's limits for each, when a spin clip is enough, and the request to make one.
- One garment on different body types with AI: use real people
Shoppers want clothes on bodies like theirs. Google's 2023 try-on showed real models XXS to 4XL. Build a range from consented photos and one Sume image call.
- Snap DPA for Sponsored Snaps: catalog images to fix with Sume
Snap's Dynamic Product Ads now run in Chat for all global advertisers. The creative comes from your catalog, so the work is catalog images Sume can fix.
Written by Sume