Remaking an existing clip: what gets billed
Four channels price work that starts from an existing video. Three run the source through a model and bill it by the second, at rates above their own text-to-video. One reads the clip into a cast, sets and a storyboard, prices that read by the minute, and never hands the source to the video model. As of 2026-10-25.
| Channel | What working from a video costs | What the source is used for |
|---|---|---|
| Luma | Video-to-video at 12 to 192 credits a second, against 4 to 80 from text | Transformed by the model |
| PixVerse | A recreation agent at 80 credits a second at 720p, 157 at 1080p | Recreated by the agent |
| Runway | A video-to-video model at 28 credits a second; Seedance 2.5 input video billed per second | Transformed, or used as reference |
| SceneMixer | A read priced by the clip's length in minutes, then ordinary drama prices | Read into a storyboard; never sent to the model |
Inclusion rule. Channels whose price list or guide prices generation that starts from a supplied video. Editing tools that only trim or caption a video are not given a row. Order. Alphabetical by channel.
1Starting from video costs more than starting from text
On one platform's own model, video-to-video runs at two and a half to three times the rate of text or image to video: 12 against 4 at draft, 192 against 80 at 1080p. A recreation agent on another platform prices a second at about nine times that platform's silent 720p rate. The source is extra input, and the price lists treat it as extra work.
2Reading a clip is a different operation
One drama tool starts the other way round. It watches the clip shot by shot, works out who is in each shot, what they do and what they say, and lays that out as a cast, locations, props and a storyboard. The read is priced by the clip's length in minutes, rounded up, and shown before starting; generating the new version is then priced like any drama.
The guide is explicit that the source is never given to the video model, so the faces that come out are the user's own cast rather than the people in the original.
3What each arrangement is good for
Transforming a source keeps its timing and camera, which suits restyling footage someone already owns. Reading it keeps only the structure, which suits taking a format that works and shooting it again with new people. The price difference follows from that: per second of source on one side, per minute of reading plus ordinary generation on the other.
- What working from a source video costsVideo-to-video on Ray3.14 is 12, 24, 48 or 192 credits a second from draft to 1080p, against 4, 10, 20 or 80 for text or image to video
- What working from a source video costsThe Viral Recreation Agent is 80 credits a second at 720p and 157 at 1080p
- Aleph 228 credits per second, with a 56-credit minimum per generationa video-to-video model
- Seedance 2.5 on the developer API20 credits per second of output at 480p and 68 at 1080p, with input and reference video billed per second on top
- What working from a source video costsReading a source clip is priced by its length in minutes, rounded up, shown before starting and charged when the read completes; everything after is priced as any drama
- What the source is used forThe source clip is never given to the video model; the faces that come out are the ones in the user's own character sheets
4Sources
Each figure is read from the vendor page linked in the ledger, checked 2026-10-25. How a figure qualifies is on method; what a rate leaves out is on limits and unspent credits and parallel generations. Nearby: Waiting time, The cheapest second, Credits vs dollars.