Core capability
The number that matters for adult work
Sixty days to 9 September 2026: 39 Seedance 2.5 reference-to-video jobs from 17 creators, 2 finished, 36 stopped by the content-safety response, 1 other failure. Median render for the two that finished was 241 seconds. Every stopped job was refunded. Set against Wan 3.0 reference-to-video in 21 days (6,901 jobs, 2.2% stopped) or Wan 2.2 Spicy (16,059 jobs, zero stopped), Seedance 2.5 is the most heavily reviewed video route on the platform for this kind of content.
What Seedance 2.5 does that nothing else here does
ByteDance documents clips up to 30 seconds, up to 50 multimodal references, precise reference control and granular editing. Studio exposes 480p or 720p, 4 to 30 seconds, seven ratios including 21:9, and reference slots for 30 images, 10 videos and 10 audio clips (40 assets total). For storyboard-driven, multi-character, non-explicit scenes with a script, it is the strongest tool on the platform, and the reason it appears in the model directory. Default quotes at review time were 1,440 credits per route.
Where the review draws its line
The two adult runs that finished were suggestive: clothed, implied, no nudity in the uploaded references. The stops cluster on references that are already explicit and on prompts with explicit verbs. Unlike Wan 3.0, which mostly rejects rendered output, the Seedance 2.5 provider refuses the request, so the run ends quickly and you keep your credits but learn nothing from rewording. Treat “implied and clothed” as the operating ceiling for this route.
Versions and current access
Seedance 2.5 text-to-video, image-to-video and reference-to-video are available in Makeartly Studio; the CTA preselects the route when the live catalogue confirms the model ID.
Evidence
The routes to use for explicit work
Wan 3.0 reference-to-video is the like-for-like alternative: up to 10 images, 5 videos, 5 audio references, 30 seconds, audio, 2.2% review stops. For single-still explicit clips, Wan 2.7 Spicy (1080p, 15 seconds, zero stops) or Wan 2.2 Spicy (loops, zero stops). Keep Seedance 2.5 for the scripted, cinematic, non-explicit parts of a project and you will get its quality without the refusals.
Deployment filtering and adult-content boundaries
Makeartly does not add a separate mainstream adult-content refusal layer for verified adult creators; this deployment-level statement does not mean the upstream official model is uncensored or guarantee prompt acceptance. Adult workflows are for adults only. CSAM and all under-18 sexual content are prohibited. Consent and likeness authorization are required, and non-consensual intimate content, unauthorized real-person deepfakes, and sexual impersonation are prohibited.
Prompting Seedance 2.5 within its ceiling
Assign references explicitly (“Image 1 is her face, Image 2 the dress, Video 1 the camera move”), write the scene as timed beats, and keep clothing on in every reference. Use 480p and 6 to 8 seconds to check casting before a 30-second 720p render, because per-run quotes are high and a refused long run wastes the most time. The 21:9 ratio and the script-style prompting are where 2.5 shines; explicit content is where it stops.
Use cases
A scripted 30-second fashion or dance scene with three references
Seedance 2.5 reference-to-video is the best tool here and this content is inside its review ceiling.
An explicit 20-second two-character scene
Wan 3.0 reference-to-video; Seedance 2.5 refused 36 of 39 comparable attempts.
- Ask whether the scene is explicit. If yes, open Wan 3.0 or a Wan Spicy route; do not start on Seedance 2.5.
- For suggestive or non-explicit scripted scenes, prepare clothed references with one job each.
- Write the casting sheet and timed beats in the prompt and choose a ratio such as 21:9 or 9:16.
- Render 6 to 8 seconds at 480p to confirm casting, then the final at 720p and full length.
- If the provider refuses a clip, move that brief to Wan 3.0 rather than rewording it.
Limitations and decision checks
Acceptance test only with clothed references
Run a short 480p clip from clothed references as the acceptance test, compare the finished motion against the storyboard, stop at the first refusal, and choose Wan 3.0 for anything the review will not pass rather than spending more attempts on 2.5.
The 92% figure is a 60-day measurement of mostly explicit traffic; a non-explicit brief will do far better. Small sample, strong signal: 39 jobs is a small sample, but 36 refusals is not ambiguous. We re-measure before changing this page.
