Uncensored Image to Video: The Parameters That Actually Matter
Turning a still into a clip with an uncensored image to video model comes down to three things: whether it accepts your photo, how long the output runs, and what a second costs.
From $0.07 per second, 480p with a reference clip
What uncensored image to video means in practice
An uncensored image to video run takes a photograph you supply and animates it without rejecting the subject. Seedance 2.5 accepts reference material containing real people, which is the specific behaviour people are looking for.
Most failed uncensored image to video attempts never reach generation. The host detects a face in the reference and refuses before anything renders, which is why the process feels broken rather than restricted.
Reference images and videos may contain real people and are checked automatically. For uncensored image to video work this is the difference between a workflow and a dead end.
You can supply up to 30 reference photos in a single uncensored image to video request. Feed a character sheet and the output holds that person across shots instead of drifting.
The platform exposes content_filter, true by default. Setting it false routes to a less restrictive channel. Their documentation calls it routing rather than an off switch, so uncensored image to video here still passes policy checks.
Speech, effects and music generate alongside the picture. An uncensored image to video clip arrives as a finished MP4 with audio inside, not a silent file waiting for a second pass.
No generator here. Uncensored image to video runs happen on the hosted API, billed per second. Everything below is checked against published parameters.
What makes Seedance 2.5 workable for uncensored image to video
Four behaviours that decide whether a photo becomes a clip.
Real faces accepted
The reference photo may contain a real person. Every uncensored image to video attempt on Higgsfield, Krea or Dreamina dies here instead.
First and last frame
Supply two stills and the model animates the transition between them. This is uncensored image to video with a defined start and end rather than a guess.
Thirty seconds of output
One uncensored image to video generation runs to 30 seconds. Seedance 2.0 and MiniMax H3 both stop at 15.
Free reference images
Reference photographs add nothing to the bill. Only output seconds and reference video length are charged.
Four steps for uncensored image to video
- 1
Host the photo
Every uncensored image to video request needs public HTTP or HTTPS image URLs. Base64 and data URIs are rejected before the job starts.
- 2
Say what moves
Describe the motion and the camera. Quoted lines become generated speech, so uncensored image to video output can talk without a separate audio step.
- 3
Choose length and size
Four to thirty seconds at 480p, 720p or 1080p. No uncensored image to video tier reaches 4K on this model.
- 4
Submit and poll
Two to five minutes per job. Polling is free and a failed run is refunded in full.
Uncensored image to video, answered
Yes. Reference images and reference videos may contain real people, reviewed automatically rather than blocked on detection. This is the single behaviour that separates a working uncensored image to video pipeline from one that refuses at upload.
No hosted model is. Moderation runs on every host, and the platform says so plainly in its own documentation. What you get is a permissive stance on real-person references plus a content_filter flag that routes to a less restrictive channel, not a switch that removes review. Treat any site promising a fully uncensored image to video service as advertising rather than a specification, because the host carries the legal exposure and prices accordingly.
Up to 30 reference images, and they cost nothing extra. Supplying several of the same subject keeps an uncensored image to video result consistent across shots, which a single frame cannot do.
Per second of finished video, from $0.071153 to $0.461856 by resolution and whether a reference video was supplied. A ten-second 1080p clip runs about $4.62. Failed jobs are refunded in full.
Check whether a task was created at all. A synchronous 400 means a schema or format problem you can fix, most often an unsupported file type or a URL that is not publicly reachable. A task that fails later returned error 80006, which is moderation and deliberately does not name the stage that fired. It could be the photo, the prompt, or the rendered output, so an uncensored image to video refusal needs the whole request re-examined rather than one field.
No. The model tops out at 1080p on this API and every other host. Any page advertising 4K uncensored image to video is quoting a specification that does not exist for this model.
Named real celebrities, third-party intellectual property and illegal content. These are enforced inside the model, so no uncensored image to video setting on any host removes them, and a rejected job is refunded rather than partially delivered.
JPEG, PNG and WebP are the safe choices, each under 30 MB and served over a public URL. The check reads the file itself rather than the extension, so renaming a file does not help an uncensored image to video request pass. Reference clips must be MP4 or MOV, between 2 and 30 seconds each.
Run uncensored image to video on the API
Real-person references accepted, up to 30 photos per request, per-second pricing and refunds when an uncensored image to video job fails.
Related: Seedance 2.5 API