Which image tier for your sheets

Quick answer: Start on the default, Qwen-Image 3.0, for everything — it is the only tier the free sheets are drawn on. Move the faces that carry the series to Qwen-Image 3.0 Pro, or to Nano Banana Pro on English-language projects with detailed costumes. Use SeedDream 5.0 Lite for 2K locations on a budget and SeedDream 5.0 Pro for photographic skin. For state images and the promo cover, which are drawn from an existing sheet, GPT Image 2.5 Sunburst is built to hold the reference. In SceneMixer, the tier choice is one setting and applies to new images only, so you can switch per group.
Still from “Starfall Command”, an AI short drama generated end-to-end by SceneMixer (7 characters, 1 episode)
SampleStarfall CommandDigital HD1 episodeProduced end-to-end by SceneMixer from the scriptWatch the episodes

The default tier has the full explanation of the two reading habits every choice here turns on: Qwen-Image 3.0.

The table

Ten image tiers can be picked in Settings → Model Preferences → Image Model. Prices are per image.

You needPickPer imageWhy
A whole cast, cheaply; the free sheetsQwen-Image 3.0 (default)7Follows plain descriptions; the only tier the free sheets are drawn on
Leads that must hold up in close-upQwen-Image 3.0 Pro18More identity in the face; marked Recommended in Settings; Chinese descriptions
2K locations and props on a budgetSeedDream 5.0 Lite82K, reads the description to the end, low price
2K and photographic skinSeedDream 5.0 Pro25Reads the description end to end; strongest skin
English supporting cast and places, a step up from the defaultNano Banana 2.119About 2K; Google’s faster model; cheapest of the new tiers
Detailed costumes and faces from long English descriptionsNano Banana Pro46About 2K; keeps the most named detail; the most expensive tier
Volume at 2K from another makerGPT Image 2.5 Flare25About 2K; OpenAI’s fast general-purpose model
State images and covers drawn from an existing sheetGPT Image 2.5 Sunburst25About 2K; OpenAI’s precise model, holds a reference subject
A cover with the title painted inGrok Imagine Image 2.035About 2K; xAI’s model, strong on lettering; very long descriptions are shortened
A different look for facesMAI-Image-2.631About 1 megapixel; Microsoft’s model; strict content filter

Free sheets: every SceneMixer account’s 15 free sheets are drawn on the default Qwen-Image 3.0 only. Pick another tier and those images are charged at that tier’s price without touching the allowance; switch back and whatever is left is still there. See free sheets are per account.

Two habits that decide most cases

How much of the description is followed. The Qwen tiers follow the start more reliably than the end; the SeedDream tiers read to the end. Character descriptions on this site are long, with identity first, which is why the Qwen tiers still do well on a cast: what matters most comes first. The same order — identity first, decoration last — also protects you on Grok Imagine, where a very long description is shortened at the end.

How literally phrasing is followed. All tiers act on what you describe; none reliably follows what you tell it not to draw. The SeedDream tiers follow result-style phrases most literally, so describe causes. See say what is there, not what is not.

A working pattern

  1. Generate the whole cast on the default tier, using the free sheets. This is where you get the wording right.
  2. Look at the leads. Switch to Qwen-Image 3.0 Pro, Nano Banana Pro or SeedDream 5.0 Pro and regenerate only those.
  3. Switch to SeedDream 5.0 Lite or GPT Image 2.5 Flare for the locations if you want 2K plates.
  4. For state images and the cover, switch to GPT Image 2.5 Sunburst (or Grok Imagine Image 2.0 for a cover with a lot of lettering).
  5. Switch back. Each switch on SceneMixer affects new images only.

What not to do

Do not iterate on a description at a premium rate: get the wording right on the default tier, then regenerate the keepers. Do not regenerate a whole cast on a premium tier when only the leads are seen in close-up. Do not expect a tier switch to redraw what you have — regenerating is a new image and a new charge.

Each tier on its own page

FAQ

Which tier is the default?

Qwen-Image 3.0.

Which tier has the Recommended badge?

Qwen-Image 3.0 Pro.

Which tiers render 2K?

SeedDream 5.0 Pro and Lite, Nano Banana 2.1 and Pro, GPT Image 2.5 Sunburst and Flare, and Grok Imagine Image 2.0. SceneMixer's MAI-Image-2.6 draws at about one megapixel.

Do the free sheets depend on the tier?

Yes. The 15 free sheets per account are drawn on the default Qwen-Image 3.0 only; on any other SceneMixer tier images are charged from the first one.

Which tiers read English descriptions best?

The Nano Banana tiers, Pro for detail and 2.1 for a lower price. Chinese descriptions read better on the Qwen and SeedDream tiers.

Which tier for the promo cover?

On SceneMixer, the cover is drawn on your selected tier. GPT Image 2.5 Sunburst holds your leads' sheets well; Grok Imagine Image 2.0 is the one to try for lettering.

Can I use different tiers for characters and locations?

Yes, by switching SceneMixer's preference between the two groups.

Where are the Gemini 3.1 Flash and Gemini 3 Pro tiers?

The Nano Banana tiers you can pick on SceneMixer now are Nano Banana 2.1 and Nano Banana Pro. Gemini 3 Pro Image is the model behind Nano Banana Pro.

Does switching tier change my existing sheets?

No. On SceneMixer, only new and regenerated images use the new tier.

Generate the cast, then upgrade the leads

Read a script, generate on the default tier with the free sheets, and move only the faces that matter to a better tier.

Start a project

Updated November 15, 2026