AISPHow to keep characters consistent across AI video episodes
Write one fixed visual description per character, generate a reference portrait and seed shots with it, lock a single visual style for the whole series, and carry each scene’s final frame into the next.
Why characters drift
An image or video model has no memory. Each shot is drawn from its prompt, so “Ada, a detective” becomes a different woman every time. Vague descriptions make it worse, and so does changing the style partway through: a character drawn as anime in one scene and photoreal in the next is effectively two people.
Drift is also cumulative. Small changes in lighting, palette and location between consecutive scenes make a sequence feel like unrelated clips even when the faces are close.
Four things that hold a series together
- A locked description per character. Write age, build, hair, clothing and distinguishing features as concrete facts, and reuse exactly the same words in every shot. Never “as described above”.
- A reference portrait. Generate one image per character and use it to seed shots where that character appears alone.
- One visual style for the whole series. Pick it once and append it to every shot prompt.
- Scene chaining. Start each scene from the final frame of the one before, so light, palette and location carry through.
How AISP applies them
AISP does all four automatically. The series bible stores each character’s appearance as a concrete description. Each character gets a reference portrait (shown in the studio as “Keeps faces consistent”), and it seeds any shot where exactly one known character appears. AISP deliberately does not seed a two-person shot with one face, because that paints the same face onto everyone.
The visual style (Cinematic, 3D Animation, Cartoon or Anime) is locked once the series is created. Each rendered scene seeds the next, and each episode opens from the frame the previous one ended on. After every episode the agent writes continuity notes, so later episodes remember what earlier ones established.
On the video tiers you can also choose Continuous instead of Scene by scene: each scene carries on from the previous scene’s footage, so motion and light run straight through the cut. It costs more per second, and the rate is shown on the button before you pick it.
What it will not fix
Group shots are the weak spot of every current model: with several people in frame, faces are held by the written descriptions rather than a portrait, so expect more variation there. If a portrait fails to generate, AISP tells you which character has none and lets you regenerate it from the series page.
Questions
Can I upload a photo to use as my character?
Not as a likeness. AISP builds characters from written descriptions and the portraits it generates; attached images are not used as face references.
Does changing the brush break consistency?
It can. The style chooses a matching brush by default, and AISP warns you if you pick a brush that cannot paint the medium your style promises.
Product details checked October 2026. Rates and features can change; the app always shows the current cost before you confirm.