“Which Model Is Best?” There Isn’t One — Pick the Tier by What the Shot Needs

Quick answer: There is no single best. Video tiers differ in resolution, maximum segment length and price per second; image tiers in how they read a description and how sharp a face they draw; text tiers in how carefully they structure a storyboard and what they cost. For most projects the defaults are the right call: Wan 3.0 · 480P at 10 credits a second for drafts, Wan 3.0 · 720P for finals, the default image and text tiers everywhere else. Switch tiers only for a reason listed below.
Still from “Crimson Vow”, an AI short drama generated end-to-end by SceneMixer (4 characters, 1 episode)
Sample: “Crimson Vow” — Realistic HD, 1 episode, produced end-to-end by SceneMixer from the script · watch the episodes

Open the model selector in the editor’s top bar: every video tier shows its resolution and its longest segment; the price per second is next to each tier in Settings.

Video: four families

Four families, each in two or three resolutions. Wan 3.0 (480P / 720P / 1080P) is the default family: segments up to 30 seconds, dialogue and sound generated with the picture; its 480P tier is the usual way to draft. Seedance 2.5 (480P / Lite / Pro) also takes 30-second segments, at a higher price per second in its upper tiers. Seedance 2.0 (480P / Fast / standard) takes 15-second segments. MiniMax H3 (H3 / H3 Fast) takes 15-second segments, and has no 480P. Wan 3.0 and H3 both cut between shots inside one segment when the storyboard writes the cuts. Every price is per second; a 12-second segment on the default tier costs 12 × 10 credits. Full list with maximum lengths: which tiers accept 30 seconds.

The SceneMixer episode editor: asset library on the left, the storyboard in the middle, the video preview on the right, and every segment along the bottom timeline
The model selector in the editor's top bar lists every video tier with its resolution and longest segment.

Video: pick by purpose

Drafting an episode: the default 480P tier. Generate every segment once, fix the storyboard where it is wrong, then regenerate only the segments you keep at a higher resolution — the draft-then-final workflow. A final for phones: 720P is the recommended tier; 1080P for landscape or a screen larger than a phone. A long single take (a speech, a slow reveal): a 30-second family. A line the default tier keeps mispronouncing: regenerate that segment on MiniMax H3. Anime and stylised looks: any family; the visual style preset does the work, and 480P hides less than you would think.

Image: two questions

Sheets are references, so the questions are: does the tier draw the face you described, and does it keep the background plain? The default Qwen-Image 3.0 reads Chinese and English descriptions well; Qwen-Image 3.0 Pro is the recommended step up for lead characters. SeedDream 5.0 Pro reads a long description in full and draws skin and cloth with a more photographic feel; SeedDream Lite is its lighter sibling. The price per sheet shows on the button before you generate. Writing the description so any tier draws the right face: Write a Description That Works.

Text: the tiers that parse and write storyboards

Text tiers read your script and write the storyboard. The default Doubao Seed 2.1 Turbo handles most scripts; Doubao Seed Evolving is the recommended step up when a storyboard keeps coming back thin or mis-structured; Doubao Seed 2.1 Pro and Qwen3.8 Max are for long, dense material; Qwen3.8 Flash is a fast option for simple scripts. The parse and storyboard fees scale with the tier’s multiplier; each tier’s rate is shown next to it in Settings. Change the text tier in Settings; to use a different tier for one episode, switch before generating that episode’s storyboard and switch back after.

How to change a tier, and what it does not change

Video tier: the button in the editor’s top bar, which also changes the video tier in Settings; each segment keeps the tier it was made with. Image and text tiers: Settings. Changing a tier never regenerates anything by itself; existing sheets and clips stay. A tier that is switched off is replaced automatically, by the default or, for video, by the tier the episode last used. Settings, Item by Item and How Model Tiers Work.

You wantPickWhy
A full draftWan 3.0 · 480P10/s, 30-second segments
Final for phonesWan 3.0 · 720PRecommended; 19/s
Final for big screensWan 3.0 · 1080P or Seedance 2.5 Pro1080P; 38/s and 75/s
Lead character sheetsQwen-Image 3.0 ProRecommended image tier
Storyboards that keep coming back thinDoubao Seed EvolvingRecommended text tier

FAQ

Which model is the best?

None outright. Draft on the default video tier, finish on 720P or 1080P, and switch only for a specific need: a 2K picture on MiniMax H3, or a line the default keeps mispronouncing (try H3 for that segment).

Does a more expensive tier keep faces more consistent?

Consistency comes from the sheets and the storyboard, not the tier. A higher tier gives sharper frames, not a steadier face.

Can I mix tiers in one episode?

Yes. The video tier is chosen per segment; the composite joins them regardless.

Can I use a different text tier for one episode?

Yes. The text tier is read each time a task starts; switch it in Settings before generating that episode’s storyboard and switch back after.

Where do I see the price before switching?

In Settings, next to each tier: per second for video, per sheet for images, as a rate for text. The generate button on a segment also shows that segment’s price.

Default to draft, upgrade to finish

The tier is a price and a resolution, not a magic wand. Spend on the segments you keep.

Open the editor

Updated September 27, 2026

Pricing·Novel to Video·Guides·Samples·Support·Privacy·Terms·Legal·Contact·© 2026 SceneMixer