Smart Shot Best Practices 2026
Smart Shot now redirects to the Long Video Generator, and that tells you what it became. Character consistency needed chaining, not a better single generation.
TL;DR
Smart Shot redirects to the Long Video Generator, because character consistency needs chaining rather than a better single generation. Three mechanisms work: reference images on thirteen models, first-to-last frame on Seedance 2.0 and 2.0 Fast, and chaining, where each segment carries the previous one's finished clip. Lock the character image first since everything inherits it, and judge each segment's final frame rather than the clip, because the next segment continues from it.
Smart Shot now takes you to the Long Video Generator. That redirect is the most useful thing to know about it, because it tells you what the tool actually became.
The problem Smart Shot was reaching for, a character who stays themselves across a sequence, turned out to need a different mechanism than a better single generation. It needed chaining, and that is what the Long Video Generator does.
Why single generations cannot solve consistency
Every generation starts from nothing. Give the same prompt twice and you get two different people, because "dark-haired woman in a grey coat" describes millions of them and the model picks one each time.
No consistency setting fixes this, and there is no consistency weight to set. What fixes it is giving the model something to look at.
The three mechanisms that do work
Reference images. Thirteen models accept a character reference through reference to video: Seedance 2.0, 2.0 Fast and 2.5, Veo 3.1 and 3.1 Lite, Wan 2.7, 3.0 and 3.0 Prime, Kling 3.0 Motion Control, MiniMax H3, Gemini Omni Flash, and both Happy Horse models. Showing beats describing.
First and last frame. Seedance 2.0 and 2.0 Fast take a start and end image and generate the movement between. Hand shot two the final frame of shot one and the join is continuous by construction.
Chaining. The strongest. Segment one renders from your references; every later segment carries the previous segment's finished clip and continues from its final frame, with character references riding along throughout. Identity and environment stay continuous because nothing is re-imagined from scratch after the first shot.
Chaining is what Smart Shot pointed at, and it runs on Seedance 2.5.
Practices that actually help
Lock the character image before anything else. Everything inherits it. Ten minutes on the reference is worth more than any amount of prompt tuning afterwards, because a mediocre reference propagates through every frame of every segment.
Reuse the identical file. Not a similar image: the same one. Two similar references produce two similar-but-different people, which is the exact failure you are avoiding.
Judge the final frame, not the clip. In a chain, the next segment continues from where this one ended. A clip that looks fine in motion but ends on a smeared or half-turned frame poisons everything after it. Check the ending specifically, before continuing.
Test one segment before committing to eight. A flaw in segment one is a flaw in all of them, and finding it at segment six is expensive.
Keep the prompt skeleton constant. Change the action; keep the character description, lighting and location wording identical. Rewriting invites reinterpretation.
Generate short. Identity holds over a few seconds and degrades over many. Several short segments cut together beat one long generation, and they fail more cheaply.
Picking the shape of the piece
One continuous run of action. The Long Video Generator and its chain.
Discrete scenes cutting between locations. The Series Generator, where each scene is generated independently and references carry the character.
A single performance transferred onto a character. Motion Control, which takes a driving video and a character image.
Choosing wrong is the costliest mistake here, because you find out after rendering.
What is not real
No consistency weight or score. No 0.90 setting, and no published consistency percentage per model. Tables listing "consistency score" and "drift after N seconds" invented both columns.
No motion strength. Across all 37 video generation models the parameters are prompt, duration, aspect ratio and resolution, plus references, a negative prompt or first and last frames on some.
No minimum reference resolution. You are not required to supply 1536x1536, and larger references do not unlock longer clips.
No frame rate control.
One model exposes a seed: Wan 2.7, along with a negative prompt. If you want the same camera behaviour across shots rather than independent interpretations, that is the only entry offering it.
An order that works
- Build one strong character reference. Frontal or three-quarter, evenly lit, sharp, large in frame.
- Decide the shape: continuous run, or discrete scenes.
- Generate segment one. Judge its final frame honestly.
- Chain the rest, checking each ending before continuing.
- Finish with Lip Sync if there is dialogue, then captions.
Steps 1 and 3 decide the result. Everything else is execution.
The catalog is at Models, each generation is quoted before it runs, and pack prices are on the pricing page.
Frequently Asked Questions
Where did Smart Shot go?▾
It redirects to the Long Video Generator. The problem it was reaching for, a character who stays themselves across a sequence, turned out to need chaining rather than a better single generation, and chaining is what the Long Video Generator does.
What consistency weight should I set?▾
There is no consistency weight, and no published consistency score per model. Tables listing values like 0.90 or a drift-after-N-seconds column invented them. Consistency comes from giving the model something to look at, not from a setting.
What are the mechanisms that actually keep a character consistent?▾
Reference images, supported by thirteen models. First-to-last frame on Seedance 2.0 and 2.0 Fast, where shot two starts on shot one's final frame. And chaining, where each segment carries the previous segment's finished clip and continues from its final frame, which holds the environment as well as the character.
What is the single highest-value habit?▾
Lock the character reference image before generating anything else. Everything inherits it, so ten minutes spent on the reference is worth more than any amount of prompt tuning afterwards. Then reuse the identical file rather than a similar one.
Why does my sequence drift after a few segments?▾
Usually because an early segment ended on a bad frame. In a chain the next segment continues from the previous one's final frame, so a clip that looks fine in motion but ends smeared or half-turned poisons everything after it. Judge the ending frame specifically before continuing.
Should I use the Long Video Generator or the Series Generator?▾
The Long Video Generator for one continuous run of action, since chaining lets each segment continue from the last. The Series Generator for discrete scenes cutting between locations, where each scene is generated independently and references carry the character.
Does reference image size affect clip length or quality?▾
There is no minimum reference resolution, and a larger reference does not unlock longer clips. What matters is that the reference is frontal or three-quarter, evenly lit, sharp and large enough in frame to carry facial detail.
