Why we don't use start and end frames
When you give Seedance 2.5 or Wan 3.0 a start frame and an end frame, they stop using your character sheets and voice samples. The clip lands on your pictures, but your characters can stop looking and sounding like themselves. So ShowMaker never uses frames: every clip gets the character sheets and voices as references, and the clip before it for the join.
29 October 2026 4 min read
Most video models let you send a first frame and a last frame, and the model fills in the motion between them. For a series that sounds perfect: draw the opening, draw the ending, and the clip has to hit both. We tested it on three models.
The test
We took one moment from our demo show Power Couple: Brett, at Issac's table, asks him for some peer feedback. We drew a start picture and an end picture, and sent each model the same two pictures as its first and last frame.
Seedance and Wan also got each character's sheet and a sample of Brett's voice. MiniMax H3 got the pictures only, since its frames mode takes nothing else. Every clip was 10 seconds at 480p.
Frames do land
All three clips start on the first picture and end on the last one. If an exact opening and an exact ending were all you needed, you could stop here.
What you lose
Look at the Seedance clip at 2.5 seconds. A second Issac is standing behind Brett.
We think we know why. Alibaba's Wan 3.0 API reference says start and end frames and references are mutually exclusive. For Seedance 2.5, OpenRouter's Seedance 2.5 review says frame images take priority, so references have no visible effect. If that is right, the faces in our clips looked right only because the pictures already showed them, and once the camera moves past what the pictures show, nothing tells the model who Issac is.
Pitfall
Nothing fails. The request is accepted, the clip comes back, and you pay for it. You get no warning that your character sheets and voice samples may not have been used.
The voice goes the same way, because a voice sample is a reference too. Neither Seedance 2.5 nor Wan 3.0 takes an audio file in frames mode. One model that does is Wan 2.7, but it is older and only makes 720p and up.
We also tried fixing the voice afterwards. We kept the frames clip and ran its sound through a voice changer into the character's voice. On the Seedance clip the pauses stayed within 0.03 seconds of the original, so the lips still matched. Two speakers worked too, but only after cutting out every line and changing each one on its own, and people talking over each other would be harder still. We dropped it: too much fiddly work on every clip of every episode.
Pictures as references instead
So we tried the other way round, on another moment from the same show, with Issac alone in the lobby. The start and end pictures go in as references, next to Issac's sheet and a British man's voice sample. The prompt says the clip opens on the first picture and ends on the second. The British voice came through.
Both clips open in the same lobby as the start picture. The ending doesn't match the end picture: both end on a closer shot of Issac, not on the full-body picture.
That is the trade we found. Frames give you an exact ending and can lose the characters. References keep the characters and give you an ending that is only roughly like the picture. For a show, the characters matter more.
What ShowMaker does
Every clip ShowMaker shoots gets each character's sheet as a reference, their voice when they have one, and the places and props it shows. When a clip continues from the one before, it also gets the last few seconds of that clip, so the join is smooth without a hard end frame.